Meteorological data downscaling method and device based on space-time diagram neural network
Through the meteorological data reduction method based on the spatio-temporal map neural network, the problem that the existing technology is difficult to accurately capture the details of meteorological phenomena in local areas is solved, and high-resolution and high-precision meteorological data reduction scale is achieved, which is suitable for different regions and time scales. meteorological data processing.
Patent Information
- Application Number
- CN202410967717.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-18
AI Technical Summary
The existing meteorological data reduction method is difficult to accurately capture the details of meteorological phenomena in local areas, which affects the resolution and accuracy of meteorological data after downscale.
Using the meteorological data reduction method based on the spatiotemporal map neural network, data preprocessing and feature variable determination are carried out by acquiring meteorological observation data and grid data of meteorological sites, and using the trained spatiotemporal map neural network to low-resolution meteorological network The grid data is processed downscale to generate high-resolution meteorological grid data, and the deviation correction is made through the second spatio-temporal map neural network to improve the detailed performance and accuracy of meteorological data.
It significantly improves the spatial and temporal resolution of meteorological data, enhances the detailed performance and accuracy of meteorological data, can more accurately capture the details of meteorological phenomena in local areas, improves the resolution and accuracy of meteorological data after downscale, and has more High generalization ability.
Smart Images

Figure CN118915192A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of meteorological data downscaling, and in particular to a meteorological data downscaling method and device based on a spatiotemporal graph neural network. Background Art
[0002] In meteorological research, high-resolution meteorological data is crucial to fields such as weather forecasting, climate change analysis, and agricultural production. Existing meteorological data usually come from meteorological station observations, satellite remote sensing, and reanalysis products, with low spatial resolution, usually within tens to hundreds of kilometers. In order to meet the needs of regional and local research, downscaling technology has become an important means to improve the spatial resolution of meteorological data.
[0003] At present, meteorological data downscaling mainly includes three types of methods: dynamic downscaling, statistical downscaling and coupled downscaling. Among them, dynamic downscaling captures the physical process of meteorological variables through numerical simulation and generates high-resolution meteorological data, but it has high computational complexity, is time-consuming, and is easily affected by input data and numerical patterns; statistical downscaling establishes a mapping relationship between high-resolution and low-resolution meteorological data through statistical models, has low computational cost and is easy to apply, but relies on long-term observation data with high temporal and spatial resolution; coupled downscaling combines the advantages of dynamic downscaling and statistical downscaling methods to provide high-precision and physically consistent results, but still faces problems such as computational complexity and data dependence, and is less used.
[0004] In statistical downscaling, convolutional neural networks (CNN), recurrent neural networks (RNN) and generative adversarial networks (GAN) have been successfully applied to image super-resolution tasks and gradually introduced into the field of meteorological data downscaling. However, the current application of these network models in meteorological data downscaling is still insufficient, and it is difficult to accurately capture the details of meteorological phenomena in local areas, which affects the resolution and accuracy of downscaled meteorological data. Summary of the invention
[0005] The purpose of this application is to provide a meteorological data downscaling method, device, electronic device and storage medium based on a spatiotemporal graph neural network to more accurately capture the details of meteorological phenomena in local areas and improve the resolution and accuracy of downscaled meteorological data.
[0006] In order to achieve the above objectives, in a first aspect, the present application provides a meteorological data downscaling method based on a spatiotemporal graph neural network, comprising: Acquire meteorological observation data sets and gridded data sets of meteorological stations in the study area, wherein the gridded data sets include low-resolution meteorological grid data, high-resolution meteorological grid data, and high-resolution land static grid data; Performing data preprocessing on the meteorological observation data set and the gridded data set, and determining corresponding characteristic variables using a preset correlation analysis algorithm; Constructing a downscaling dataset according to the meteorological observation dataset, the gridded dataset and corresponding characteristic variables; The data set for downscaling is input into a trained first spatiotemporal graph neural network, and the low-resolution meteorological grid data is downscaled using the first spatiotemporal graph neural network to obtain predicted high-resolution meteorological grid data.
[0007] In a preferred embodiment of the present application, the step of obtaining a meteorological observation data set and a gridded data set of meteorological stations in a study area includes: Obtain meteorological observation data sets from meteorological stations in the study area, as well as low-resolution meteorological data and high-resolution land static data; Gridding the research area to obtain corresponding grids; According to the divided grid, the low-resolution meteorological data and the high-resolution land static data, low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data are obtained, and the low-resolution meteorological grid data, the high-resolution meteorological grid data and the high-resolution land static grid data are used as gridded data sets.
[0008] In a preferred embodiment of the present application, the low-resolution meteorological grid data, the high-resolution land static grid data and the high-resolution meteorological grid data are obtained according to the divided grid, the low-resolution meteorological data and the high-resolution land static data, including: Interpolating the low-resolution meteorological data to the divided grids to obtain low-resolution meteorological grid data; Based on the divided grids and the high-resolution land static data, high-resolution land static grid data are calculated; the high-resolution land static grid data and the low-resolution meteorological grid data are input into a preset meteorological research and forecasting modeling system to obtain high-resolution meteorological grid data.
[0009] In a preferred embodiment of the present application, the data set for downscaling is constructed based on the meteorological observation data set, the gridded data set and the corresponding characteristic variables, including: According to the determined number of lag steps, the time lag variable corresponding to the current time is obtained; According to the meteorological observation dataset, the gridded dataset, the corresponding characteristic variables and the corresponding time lag variables, a lag time characteristic dataset is constructed as a dataset for downscaling.
[0010] In a preferred embodiment of the present application, the first spatiotemporal graph neural network includes an embedding layer, a graph structure building module, a spatiotemporal convolution module, an attention module and an output layer. The embedding layer of the first spatiotemporal graph neural network has the functions of spatiotemporal coding, average coding and feature embedding, and the output layer of the first spatiotemporal graph neural network includes a two-dimensional convolutional layer and a fully connected layer in the time and space dimensions.
[0011] In a preferred embodiment of the present application, after inputting the data set for downscaling into a trained first spatiotemporal graph neural network, and downscaling the low-resolution meteorological grid data using the first spatiotemporal graph neural network to obtain predicted high-resolution meteorological grid data, the method further includes: The data set used for downscaling and the predicted high-resolution meteorological grid data are input into a trained second spatiotemporal graph neural network, and the predicted high-resolution meteorological grid data is corrected for deviation using the second spatiotemporal graph neural network to obtain corrected high-resolution meteorological grid data.
[0012] In a preferred embodiment of the present application, the second spatiotemporal graph neural network includes a data processing module, a spatiotemporal representation learning module, an uncertainty modeling module and an extrapolation generation module.
[0013] In a second aspect, the present application provides a meteorological data downscaling device based on a spatiotemporal graph neural network, comprising: a data acquisition unit, used to acquire a meteorological observation data set and a gridded data set of a meteorological station in a study area, wherein the gridded data set includes low-resolution meteorological grid data, high-resolution meteorological grid data, and high-resolution land static grid data; a data preprocessing unit, used to perform data preprocessing on the meteorological observation data set and the gridded data set, and determine corresponding characteristic variables using a preset correlation analysis algorithm; A data set construction unit, used to construct a data set for downscaling according to the meteorological observation data set, the gridded data set and corresponding characteristic variables; A data downscaling unit is used to input the data set for downscaling into a trained first spatiotemporal graph neural network, and use the first spatiotemporal graph neural network to downscale the low-resolution meteorological grid data to obtain predicted high-resolution meteorological grid data.
[0014] In a third aspect, the present application provides an electronic device comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned meteorological data downscaling method based on spatiotemporal graph neural network.
[0015] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned meteorological data downscaling method based on spatiotemporal graph neural network.
[0016] This application proposes a meteorological data downscaling method, device, electronic device and storage medium based on spatiotemporal graph neural network. Compared with the prior art, its beneficial effects are: The present application obtains meteorological observation data sets and gridded data sets of meteorological stations in the study area, wherein the gridded data sets include low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data, and uses a preset correlation analysis algorithm to determine the corresponding characteristic variables, thereby improving the diversity and richness of data sources, covering different resolutions and data types, and improving the prediction accuracy of the neural network model; at the same time, the trained first spatiotemporal graph neural network is used to downscale the low-resolution meteorological grid data, so that the predicted high-resolution meteorological grid data can be obtained more quickly, significantly improving efficiency, and the first spatiotemporal graph neural network can provide higher spatial and temporal resolution, which can enhance the detail expression and accuracy of meteorological data, thereby more accurately capturing the details of meteorological phenomena in local areas, and improving the resolution and accuracy of downscaled meteorological data; in addition, the first spatiotemporal graph neural network also has higher generalization ability, and can also be used to process different local areas, different time scales or different meteorological types. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0018] Figure 1 It is a flow chart of a meteorological data downscaling method based on a spatiotemporal graph neural network provided in an embodiment of the present application; Figure 2 It is a schematic diagram of the structure of the first spatiotemporal graph neural network provided in an embodiment of the present application; Figure 3 It is a schematic diagram of the downscaling result of the temperature at a height of 2 m above the ground and its prediction error provided by an embodiment of the present application; Figure 4 It is a schematic diagram of the structure of the second spatiotemporal graph neural network provided in an embodiment of the present application; Figure 5 It is a structural block diagram of a meteorological data downscaling device based on a spatiotemporal graph neural network provided in an embodiment of the present application; Figure 6 It is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0019] The specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present application but are not intended to limit the scope of the present application.
[0020] In response to the above-mentioned problems in the prior art, the embodiments of the present application provide a meteorological data downscaling method, device, electronic device and storage medium based on a spatiotemporal graph neural network, which can provide higher spatial and temporal resolution, enhance the detail expression and accuracy of meteorological data, thereby more accurately capturing the details of meteorological phenomena in local areas and improving the resolution and accuracy of downscaled meteorological data.
[0021] See also Figure 1 , Figure 1 It is a flow chart of a meteorological data downscaling method based on a spatiotemporal graph neural network provided in an embodiment of the present application.
[0022] In an embodiment of the present application, the following meteorological data downscaling method based on spatiotemporal graph neural network can be applied to computer devices such as servers.
[0023] In one embodiment, the present application provides a meteorological data downscaling method based on a spatiotemporal graph neural network, comprising the following steps: Step S110, obtaining meteorological observation data sets and gridded data sets of meteorological stations in the study area, wherein the gridded data sets include low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data.
[0024] In one embodiment, there are usually multiple meteorological stations in the study area, and the meteorological observation data set consists of meteorological observation data observed by the meteorological stations; optionally, the meteorological observation data may include data such as the temperature 2 m above the ground, the relative humidity 2 m above the ground, the wind speed and direction 10 m above the ground, and the accumulated rainfall.
[0025] Optionally, the low-resolution meteorological grid data and the high-resolution meteorological grid data may include data such as the air temperature 2 m above the ground, the relative humidity 2 m above the ground, the wind speed and direction 10 m above the ground, the accumulated rainfall, the surface temperature, the soil moisture and the shortwave radiation; the high-resolution land static grid data may include data such as building vectors, green space vectors, water body vectors and land cover types.
[0026] In one embodiment, after obtaining high-resolution land static grid data, the building density, building height, impermeable rate, green space rate, water body rate, terrain height and land use type in each grid can also be calculated based on the high-resolution land static grid data.
[0027] Step S120, performing data preprocessing on the meteorological observation data set and the gridded data set, and determining corresponding characteristic variables using a preset correlation analysis algorithm.
[0028] In one embodiment, data preprocessing can be performed on the data of the meteorological observation data set and the gridded data set by eliminating invalid data and filling missing data, wherein eliminating invalid data can be deleting data records containing obvious errors or abnormal values, for example, by statistical analysis or setting thresholds to identify and eliminate data points that are beyond a reasonable range; filling missing data can adopt appropriate methods to process missing values, such as using the mean, median, adjacent data points or interpolation method to fill missing values, so as to ensure the integrity of the data in time series and spatial distribution.
[0029] In one embodiment, after invalid data is eliminated and missing data is filled, the continuous data may be standardized to convert data of different scales to a unified scale and eliminate the influence between different dimensions; optionally, the data standardization method may adopt Min-Max standardization or Z-Score standardization.
[0030] By strictly preprocessing the data in the above way, data noise and missing problems can be significantly reduced, thereby improving the quality and reliability of the data.
[0031] In one embodiment, the corresponding feature variables determined by a preset correlation analysis algorithm are used as input data of the trained first spatiotemporal graph neural network, and thus the corresponding feature variables may be variables with the highest correlation with the high-resolution meteorological grid data, including the air temperature at 2m above the ground, surface temperature, shortwave radiation in the low-resolution meteorological grid data, and high-resolution land static grid data.
[0032] Step S130, constructing a data set for downscaling based on the meteorological observation data set, the gridded data set and the corresponding characteristic variables.
[0033] Step S140: input the data set for downscaling into the trained first spatiotemporal graph neural network, and use the first spatiotemporal graph neural network to downscale the low-resolution meteorological grid data to obtain predicted high-resolution meteorological grid data.
[0034] Specifically, a data set for downscaling is input into a trained first spatiotemporal graph neural network, and a gridded data set and corresponding feature variables in the data set for downscaling are input into the trained first spatiotemporal graph neural network as data sets to be predicted by downscaling.
[0035] In one embodiment, the first spatiotemporal graph neural network may include an embedding layer, a graph structure building module, a spatiotemporal convolution module, an attention module, and an output layer. For details, see Figure 2 , Figure 2 The structure and processing process of the first spatiotemporal graph neural network are shown. The first spatiotemporal graph neural network can be obtained by training the spatiotemporal graph neural network to be trained through the constructed historical meteorological data set. When training the spatiotemporal graph neural network to be trained, the historical meteorological data set can be divided into a training set, a validation set, and a test set for training.
[0036] Among them, the embedding layer of the first spatiotemporal graph neural network is used to map the input meteorological data and related features into a high-dimensional space to form an embedding vector, so that the subsequent network modules can efficiently process different types of data. The embedding layer includes the functions of spatiotemporal coding, average coding and feature embedding. For the spatiotemporal coding part, in order to enhance the spatiotemporal characteristics of meteorological data, the spatiotemporal coding can be explicitly introduced into the input features of each latitude and longitude grid as its auxiliary feature. The specific auxiliary feature f st as follows: s encode =[sinh,cos h,sinw,cos w,sinh cos w,sinh sinw] f st =[t encode ,s encode ] Among them, t represents the current time, s represents the current position, h represents the latitude of the grid, and w represents the longitude of the grid; For the average coding part, since the land use types calculated using high-resolution static land grid data use different values to represent different land use types, in order to eliminate the impact of the numerical size of the categorical variable on the model, the average coding method is used to embed the land use type features, and then the result is used as part of the input feature; For the feature embedding part, the initial features of a data sample are enhanced with spatiotemporal coding features and converted into average coding features to obtain the input features of the model. The 1×1 two-dimensional convolution can be used to perform preliminary feature extraction and transformation on the input features to obtain a high-dimensional embedding representation X emb ∈R T,N,d, which can enhance the expressiveness and robustness of the model, where N represents the number of longitude and latitude grids, T represents the time window size used in training, d0 represents the number of input features, and d represents the number of features represented by high-dimensional embedding. The embedding process can be expressed as: X emb =BatchNorm2d(Conv2d(X)) Furthermore, the graph structure building module of the first spatiotemporal graph neural network is used to represent the node embedding X based on the embedding layer. emb Construct a graph structure and calculate the similarity between nodes based on the time series embedding representation of each node. For this purpose, the K-nearest neighbor (KNN) method can be used to construct edges between pairs of nodes with high similarity. Finally, the corresponding graph structure is dynamically generated during the model optimization process for use by the spatiotemporal graph neural network to better capture the high-order spatiotemporal dependency information between nodes. Specifically, the node embedding representation X emb After flattening, we get the node time series embedding representation X emb ′∈R N,T×d , and then use the K nearest neighbor method to construct the adjacency matrix A; each time the model is iterated and optimized, a more expressive and robust embedding representation X is obtained emb , so the graph structure A constructed subsequently changes dynamically during the training process. The specific process is as follows: (1) Node embedding representation flattened: X emb ′=reshape(X emb ,(N,T×d)) (2) Calculate the similarity between nodes: Use Euclidean distance or cosine similarity to calculate the similarity matrix S between nodes. The calculation expression is as follows: S ij =cosine_similarity(X ′ emb [i],X ′ emb [j]) Where S ij Represents the similarity between node i and node j; (3) Construct the adjacency matrix: According to the similarity matrix S and the K nearest neighbor method, the adjacency matrix A is constructed. For each node i, its edges with the first K most similar nodes are retained. The element A of the adjacency matrix ij Defined as: (4) Dynamically update graph structure: After each model iteration, the new embedding representation X embThe adjacency matrix A is dynamically updated. This process is repeated during training, so that the graph structure is gradually optimized and the high-order spatiotemporal dependency information between nodes is better captured.
[0037] Furthermore, the spatiotemporal convolution module of the first spatiotemporal graph neural network is a spatiotemporal graph neural network module, which is used to process spatiotemporal graph data with a dynamic graph structure. It combines temporal convolution and graph convolution, extracts and processes spatiotemporal features through multiple steps, thereby ensuring that the model can simultaneously process the temporal dynamic changes and spatial distribution characteristics of meteorological data; in specific implementation, the spatiotemporal feature extraction is carried out by combining the temporal convolution network (TCN) and the graph convolution network (GCN) to obtain the node spatiotemporal embedding representation H. The specific process is as follows: (1) Temporal Convolutional Network (TCN): This module uses two TCNs to capture the temporal features of the embedded representation before and after the graph convolution operation. Assuming that the convolution kernel size of TCN is k, the temporal convolution operation can be expressed as: X TCN =Conv1D(X emb ,k) (2) Graph Convolutional Network (GCN): Perform graph convolution on the features output by the first layer of TCN to capture the features in the spatial dimension. Assuming that the weight matrix of the graph convolution is W and the adjacency matrix is A, the graph convolution operation can be expressed as: X GCN =σ(AX TCN W) Among them, σ represents the activation function, such as ReLU; (3) Spatiotemporal Graph Neural Network: Finally, the entire spatiotemporal feature learning process can be expressed as: H=TCN2(GCN(TCN1(X emb ))) Furthermore, the attention module of the first spatiotemporal graph neural network is used to learn the contextual information of nodes in the spatial dimension and the associated information of the same node in the temporal dimension. This module introduces the cross-attention mechanism (Criss-CrossAttention), which weights the features of the target pixel by calculating the relationship between the target feature pixel and other pixels, thereby capturing dense contextual information while maintaining long-distance spatial dependencies. Specifically, the output H of the spatiotemporal convolution module is input into the attention module as a feature map, and then the cross-attention is used to capture the contextual information of the node in the spatial dimension and the association information of the same node in the temporal dimension, and finally the aggregated node embedding representation H′ is generated: A = softmax(f affinity (QK T )) H′ = A ⊙ V Wherein, Q and K are feature maps generated by H through a 1×1 convolutional layer. To reduce the computational complexity, let C' < C. V is a feature map with the same dimension as H and is used for feature adaptation. The affinity operation aims to calculate the relationships between each node in the graph and other nodes in the same row and column, and generate a new feature map. Then, softmax normalization is adopted to obtain a new attention map A. The ⊙ operation represents element-wise multiplication of matrices. After multiplying the attention map and V element-wise, the node embedding representation H' is obtained.
[0038] Furthermore, the output layer of the first spatio-temporal graph neural network is used to receive the node embedding representation output from the attention module and generate a predicted value of the initial high-resolution meteorological data. Specifically, the output layer may include two-dimensional convolutional layers and fully connected layers in the time and space dimensions, map the embedding representation with the time window length to the target time step, and map the high-dimensional feature back to the original space to generate the downscaled meteorological data.
[0039] In one embodiment, after the first spatio-temporal graph neural network performs downscaling on the low-resolution meteorological grid data to generate the downscaled meteorological data, post-processing can be performed on the generated initial value of the high-resolution meteorological data to obtain the predicted high-resolution meteorological grid data. The post-processing methods may include outlier detection and removal, interpolation in space and time, data smoothing and filtering in space, and data smoothing and filtering in time. Post-processing can effectively improve the quality of the downscaled meteorological data and ensure the physical consistency, spatial continuity, and temporal continuity of the data.
[0040] The structure of the first spatiotemporal graph neural network mentioned above will perform effective data processing and feature enhancement during the downscaling process, thereby enhancing the detail expression and accuracy of meteorological data, capturing the details of meteorological phenomena in local areas, and more accurately predicting high-resolution meteorological data; specifically, through the above method, the embedding layer of the first spatiotemporal graph neural network regards meteorological data and its geographic information as graph node data, uses graph neural networks to process complex spatiotemporal relationships, improves the efficiency and accuracy of data processing, and enhances feature representation capabilities through spatiotemporal encoding and average encoding to ensure the robustness and stability of the model; the graph structure construction module dynamically updates the graph structure after each model iteration, captures high-order spatiotemporal dependency information between nodes, and improves model performance. Compared with the traditional method of using convolutional networks and static graph structures, it can better capture global information in space rather than limited to local information, and adaptively adjusts the graph structure according to the optimization target, which is conducive to improving the model effect; the spatiotemporal convolution module can simultaneously learn the features of the two dimensions of time and space, process the spatiotemporal dynamic changes of meteorological data, and improve the model prediction accuracy; the attention module improves the accuracy of meteorological data prediction by simultaneously capturing the correlation information in the spatial and temporal dimensions.
[0041] The above-mentioned meteorological data downscaling method based on spatiotemporal graph neural network obtains the meteorological observation data set and grid data set of meteorological stations in the study area, among which the grid data set includes low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data, and uses the preset correlation analysis algorithm to determine the corresponding characteristic variables, which improves the diversity and richness of the data source, covers different resolutions and data types, and can improve the prediction accuracy of the neural network model; at the same time, the trained first spatiotemporal graph neural network is used to downscale the low-resolution meteorological grid data, which can obtain the predicted high-resolution meteorological grid data more quickly, significantly improving the efficiency, and the first spatiotemporal graph neural network can provide higher spatial and temporal resolution, which can enhance the detail expression and accuracy of meteorological data, thereby more accurately capturing the details of meteorological phenomena in local areas and improving the resolution and accuracy of downscaled meteorological data; in addition, the first spatiotemporal graph neural network also has higher generalization ability and can also be used to process different regions, different time periods or different meteorological types.
[0042] For this, see Figure 3 , Figure 3It is a schematic diagram of the downscaling results of the temperature at a height of 2m above the ground and its prediction error using the meteorological data downscaling method based on the spatiotemporal graph neural network according to the embodiment of the present application, wherein the high-resolution meteorological grid data is simulated using a preset meteorological research and forecasting modeling system, and the simulation process takes a long time; the predicted high-resolution meteorological grid data is obtained using the trained first spatiotemporal graph neural network, which takes a very short time, can greatly shorten the time consumption, and significantly improve the efficiency, and by comparing with the high-resolution meteorological grid data, it can be seen that the predicted high-resolution meteorological grid data can effectively restore the details of meteorological phenomena; the prediction error is the difference between the high-resolution meteorological grid data and the predicted high-resolution meteorological grid data, and it can be seen that the prediction error is low and meets the accuracy requirements.
[0043] In one embodiment, when obtaining the meteorological observation dataset and gridded dataset of the meteorological stations in the study area, the following can be done: Obtain meteorological observation data sets from meteorological stations in the study area, as well as low-resolution meteorological data and high-resolution land static data; Grid the study area to obtain the corresponding grid; According to the grid division, low-resolution meteorological data and high-resolution land static data, low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data are obtained, and the low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data are used as gridded data sets.
[0044] In this embodiment, low-resolution meteorological data may be derived from numerical weather forecast models or remote sensing observations, and high-resolution static land data may be derived from resource satellite remote sensing inversion and geographic information system data.
[0045] In this embodiment, when the study area is divided into grids, the grid spatial resolution of the study area division can be determined based on the semivariogram fitting results of different meteorological elements in the meteorological station observation data; when the meteorological station data is missing or insufficient to fit, the grid spatial resolution can be directly set to 0.5kmX0.5km, and then the corresponding division grid can be obtained through grid spatial resolution division.
[0046] In this embodiment, when low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data are obtained according to the divided grids, low-resolution meteorological data and high-resolution land static data, the low-resolution meteorological grid data can be interpolated to the divided grids to obtain low-resolution meteorological grid data; According to the grid division and high-resolution land static data, high-resolution land static grid data is calculated; The high-resolution land static grid data and the low-resolution meteorological grid data are input into a preset meteorological research and forecasting modeling system to obtain high-resolution meteorological grid data.
[0047] Specifically, bilinear interpolation or cubic convolution interpolation method can be used to interpolate low-resolution meteorological data to the divided grid to obtain low-resolution meteorological grid data; and high-resolution land static grid data and low-resolution meteorological grid data are input into a preset meteorological research and forecasting modeling system (such as WRF), and the urban canopy model is turned on during the simulation, and dynamic downscaling is performed through the WRF mode to obtain high-resolution meteorological grid data.
[0048] Through the above method, a variety of meteorological data can be effectively and accurately collected, including site meteorological data and meteorological gridded data, which ensures the diversity of data sources and covers different resolutions and data types, thereby improving the prediction accuracy of the model.
[0049] In one embodiment, when constructing a downscaling dataset based on a meteorological observation dataset, a gridded dataset, and corresponding characteristic variables, the following steps may be performed: According to the determined number of lag steps, the time lag variable corresponding to the current time is obtained; According to the meteorological observation dataset, the gridded dataset, the corresponding characteristic variables and the corresponding time lag variables, a lag time characteristic dataset is constructed as a dataset for downscaling.
[0050] In this embodiment, the values of the previous n time points (lag steps) can be introduced as the features of the current time point to form a time lag variable. The introduction of the time lag variable helps the model capture the autocorrelation of the data and understand how the past state affects the current state, so that the model has a better prediction effect. The number of lag steps n can be determined by the following methods: (1) Autocorrelation analysis and partial autocorrelation analysis: Identify significant lag steps n by plotting and analyzing autocorrelation functions and partial autocorrelation function graphs; (2) Domain knowledge: Combine professional knowledge in the field of meteorology to select a time lag step with actual physical significance. For example, the daily periodic changes in meteorological data may suggest using 24 or 12 hours as the lag step n; (3) Cross-validation: Test the model performance of different n values through the cross-validation method and select the lag step n that gives the best model prediction effect.
[0051] In one embodiment, when constructing a data set for downscaling, a moving average feature may be introduced. The moving average feature is formed by calculating the average value of data within a certain window size m. The moving average feature smoothes short-term fluctuations in the data and emphasizes long-term trends, so that the model can better capture the overall trend of the data. For a certain window size m, it can be determined according to the following methods: (1) Rolling window analysis: try rolling windows with different window sizes m and determine the optimal window size by verifying the model performance; (2) Rule of thumb: select a generally effective window size m based on domain experience and historical data characteristics. For example, the daily periodic changes in meteorological data may recommend using 6 or 3 hours as the window size m.
[0052] See also Figure 1 In one embodiment, in step S140, the data set for downscaling is input into the trained first spatiotemporal graph neural network, and the low-resolution meteorological grid data is downscaled using the first spatiotemporal graph neural network to obtain predicted high-resolution meteorological grid data. The meteorological data downscaling method based on the spatiotemporal graph neural network of the present application further includes: Step S150, input the data set used for downscaling and the predicted high-resolution meteorological grid data into the trained second spatiotemporal graph neural network, and use the second spatiotemporal graph neural network to correct the deviation of the predicted high-resolution meteorological grid data to obtain corrected high-resolution meteorological grid data.
[0053] Specifically, the dataset for downscaling is input into the trained second spatiotemporal graph neural network, and the meteorological observation dataset, high-resolution land static grid data, corresponding characteristic variables and predicted high-resolution meteorological grid data in the dataset for downscaling are input into the trained second spatiotemporal graph neural network as the dataset to be corrected for deviations.
[0054] By correcting the deviations of the predicted high-resolution meteorological grid data, the trends and detailed characteristics of meteorological changes can be reflected more accurately, and the resolution and accuracy of the downscaled meteorological data can be further improved, so that it can be more effectively applied to various practical applications such as weather forecasting, disaster warning and environmental monitoring.
[0055] In one embodiment, in step S120, data preprocessing is performed on the meteorological observation dataset and the gridded dataset, and the corresponding characteristic variables are determined using a preset correlation analysis algorithm. The corresponding characteristic variables determined include not only the variables with the highest correlation with the high-resolution meteorological grid data, but also the variables with the highest correlation with the meteorological observation data of the meteorological site, including high-resolution meteorological data prediction values, land static data, and time information, etc. The characteristic variables are used to construct a dataset for downscaling as input data for the second spatiotemporal graph neural network.
[0056] In one embodiment, the second spatiotemporal graph neural network may include a data processing module, a spatiotemporal representation learning module, an uncertainty modeling module, and an extrapolation generation module. Figure 4 , Figure 4The structure and processing process of the second spatiotemporal graph neural network are shown. The second spatiotemporal graph neural network can be obtained by training the spatiotemporal graph neural network to be trained with the constructed historical meteorological data set and the corresponding predicted high-resolution meteorological grid data. When training the spatiotemporal graph neural network to be trained, the historical meteorological data set and the corresponding predicted high-resolution meteorological grid data can be divided into a training set, a validation set and a test set for training.
[0057] Among them, the data processing module of the second spatiotemporal graph neural network is used to construct a data set and an adjacency matrix according to the grid where the meteorological station is located, specifically including: (1) The input meteorological station data is regarded as nodes in the graph, and the nodes of each time step are divided into context nodes and target nodes. Each node contains exogenous variables and predictor variables, where the exogenous variables include the predicted high-resolution meteorological grid data obtained by the first spatiotemporal graph neural network and the corresponding land use static variables, and the predictor variables are the meteorological observation data corresponding to the predicted high-resolution meteorological grid data; (2) For the missing values of the target node prediction variable, a learnable embedding vector is used to fill them; (3) For all input nodes, time features are added to their exogenous variables to achieve feature enhancement. The added time features are: Among them, t represents the current time; (4) For all input nodes, the adjacency matrix of the graph is calculated based on their latitude and longitude information, the Euclidean distance between two nodes is calculated as the weight of the edge, and the obtained adjacency matrix is normalized.
[0058] Furthermore, the spatiotemporal representation learning module is used to model the spatiotemporal dependency of node features. It combines dilated non-causal convolution and graph convolution to extract and process spatiotemporal features through multiple steps, thereby ensuring that the model can simultaneously process the temporal changes and spatial distribution characteristics of node data, including: (1) Dilated Non-Causal Convolution (DNCConv): In order to capture the long-term dependency of node features, expand the receptive field of the convolutional neural network, and make data correction utilize both historical information and future information, the spatiotemporal representation learning module uses two dilated non-causal convolutions to capture the features of the time dimension of the embedded representation before and after the graph convolution operation; assuming that the convolution kernel size of DNCConv is k and the dilation factor is d, the temporal convolution operation can be expressed as: X DNCConv =Conv1D(X emb ,k,d) Among them, X emb Represents node embedding representation; (2) Graph Convolutional Network (GCN): Perform graph convolution on the features output by the first layer of TCN to capture the features in the spatial dimension. Assuming that the weight matrix of the graph convolution is W and the adjacency matrix is A, the graph convolution operation can be expressed as: X GCN =σ(AX DNCConv W) Among them, σ represents the activation function, such as ReLU; (3) Spatiotemporal graph convolution: Finally, the entire spatiotemporal representation learning process can be expressed as: H = DNCConv2(GCN(DNCConv1(X emb ))) Furthermore, the uncertainty modeling module is used to improve the robustness and stability of the model and enhance the generalization ability of the model. It introduces uncertainty modeling of node embedding representation by using graph Bayesian aggregation and latent variable state transfer with a gating mechanism, including: (1) Graph Bayesian aggregation: The information of context nodes is aggregated through the Bayesian method, taking into account the uncertainty of each context node, so that the model can better handle the uncertainty of different nodes, thereby improving the reliability and accuracy of the extrapolation results; (2) Latent variable state transfer with gating: Based on the spatiotemporal representation obtained in the deterministic representation learning stage, the latent variables are passed to the next layer through state transfer with a gating mechanism to capture deeper spatiotemporal dependencies and transfer and aggregate uncertainty information layer by layer. By introducing the gating mechanism, the effective information of the latent variables can be retained during the state transfer process, while the redundant information can be removed, thereby improving the stability of the model.
[0059] Furthermore, the extrapolation generation module is used to extrapolate the prediction variables of the target node, that is, to predict the distribution of the prediction variables of the target node. Specifically, by combining the latent variable representations of all layers and the exogenous variables of the target node, a likelihood function is used to generate the distribution of the target node. This process is essentially a conditional generation process, in which the prediction of the target node is based on the information of its surrounding context nodes and the uncertainty learned in the model. The final extrapolated result can be sampled from the distribution, or the mean of the distribution can be directly used as the prediction value, which is the modified high-resolution meteorological grid data.
[0060] The structure of the second spatiotemporal graph neural network mentioned above can effectively capture the long-term dependency and spatial distribution characteristics of node features in the process of deviation correction, and handle the uncertainty of different nodes, so as to achieve the purpose of data deviation correction and further improve the resolution and accuracy of meteorological data after downscaling; specifically, through the above method, the spatiotemporal representation learning module combines the dilated non-causal convolution (DNCConv) and graph convolution (GCN), which can more effectively capture the long-term dependency and spatial distribution characteristics of node features in the meteorological field and improve the representation ability of the model on spatiotemporal data; the uncertainty modeling module introduces graph Bayesian aggregation and latent variable state transfer to enable the model to better handle the uncertainty of different nodes, and improve the robustness and stability of the model through the gating mechanism; the extrapolation generation module generates the distribution of the target node prediction variable by combining the latent variable representation of all layers and the exogenous variables of the target node, and can obtain the prediction value by sampling from the prediction distribution or directly using the mean of the distribution, thereby improving the flexibility and accuracy of the prediction results.
[0061] In order to execute the methods corresponding to the above embodiments and achieve corresponding functions and technical effects, a meteorological data downscaling device based on a spatiotemporal graph neural network is provided below.
[0062] See also Figure 5 , Figure 5 It is a structural block diagram of a meteorological data downscaling device based on a spatiotemporal graph neural network provided in an embodiment of the present application.
[0063] In one embodiment, the meteorological data downscaling device based on spatiotemporal graph neural network of the present application includes: The data acquisition unit 510 is used to acquire the meteorological observation data set and the grid data set of the meteorological stations in the study area, and the grid data set includes low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data; the data preprocessing unit 520 is used to perform data preprocessing on the meteorological observation data set and the grid data set, and determine the corresponding characteristic variables using a preset correlation analysis algorithm; A data set construction unit 530 is used to construct a data set for downscaling based on the meteorological observation data set, the gridded data set and the corresponding characteristic variables; The data downscaling unit 540 is used to input the data set for downscaling into the trained first spatiotemporal graph neural network, and use the first spatiotemporal graph neural network to downscale the low-resolution meteorological grid data to obtain predicted high-resolution meteorological grid data.
[0064] In one embodiment, the data acquisition unit 510 may be specifically used for: Obtain meteorological observation data sets from meteorological stations in the study area, as well as low-resolution meteorological data and high-resolution land static data; Grid the study area to obtain the corresponding grid; According to the grid division, low-resolution meteorological data and high-resolution land static data, low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data are obtained, and the low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data are used as gridded data sets.
[0065] As an optional implementation, when the data acquisition unit 510 obtains low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data according to the divided grid, low-resolution meteorological data and high-resolution land static data, it can: Interpolating low-resolution meteorological data to the divided grid to obtain low-resolution meteorological grid data; According to the grid division and high-resolution land static data, high-resolution land static grid data is calculated; The high-resolution land static grid data and the low-resolution meteorological grid data are input into a preset meteorological research and forecasting modeling system to obtain high-resolution meteorological grid data.
[0066] In one embodiment, the data set construction unit 530 may be specifically configured to: According to the determined number of lag steps, the time lag variable corresponding to the current time is obtained; According to the meteorological observation dataset, gridded dataset, corresponding characteristic variables and corresponding time lag variables, a lagged time characteristic dataset is constructed as the dataset for downscaling.
[0067] In one embodiment, the meteorological data downscaling device based on the spatiotemporal graph neural network of the present application also includes: a data deviation correction unit 550, which is used to input the data set used for downscaling and the predicted high-resolution meteorological grid data into a trained second spatiotemporal graph neural network, and use the second spatiotemporal graph neural network to perform deviation correction on the predicted high-resolution meteorological grid data to obtain corrected high-resolution meteorological grid data.
[0068] The above-mentioned meteorological data downscaling device based on spatiotemporal graph neural network can implement the above-mentioned meteorological data downscaling method based on spatiotemporal graph neural network. The specific limitations and other contents of the above-mentioned meteorological data downscaling device based on spatiotemporal graph neural network embodiment can be found in the contents of the meteorological data downscaling method based on spatiotemporal graph neural network above, and will not be repeated in the embodiment.
[0069] In one embodiment, the present application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned meteorological data downscaling method based on spatiotemporal graph neural network.
[0070] Optionally, the electronic device may be a computer device, such as a server.
[0071] In one embodiment, the internal structure of the computer device of the present application can be as follows: Figure 6 shown.
[0072] In one embodiment, the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned meteorological data downscaling method based on spatiotemporal graph neural network.
[0073] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.
[0074] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
[0075] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0076] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.
[0077] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0078] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations; at the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
Claims
1. A meteorological data downscaling method based on spatiotemporal graph neural network, characterized in that: include: Acquire meteorological observation data sets and gridded data sets of meteorological stations in the study area, wherein the gridded data sets include low-resolution meteorological grid data, high-resolution meteorological grid data, and high-resolution land static grid data; Performing data preprocessing on the meteorological observation data set and the gridded data set, and determining corresponding characteristic variables using a preset correlation analysis algorithm; Constructing a downscaling dataset according to the meteorological observation dataset, the gridded dataset and corresponding characteristic variables; The data set for downscaling is input into a trained first spatiotemporal graph neural network, and the low-resolution meteorological grid data is downscaled using the first spatiotemporal graph neural network to obtain predicted high-resolution meteorological grid data.
2. The meteorological data downscaling method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The acquisition of meteorological observation data sets and gridded data sets of meteorological stations in the study area includes: Obtain meteorological observation data sets from meteorological stations in the study area, as well as low-resolution meteorological data and high-resolution land static data; Gridding the research area to obtain corresponding grids; According to the divided grid, the low-resolution meteorological data and the high-resolution land static data, low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data are obtained, and the low-resolution meteorological grid data, the high-resolution meteorological grid data and the high-resolution land static grid data are used as gridded data sets.
3. The meteorological data downscaling method based on spatiotemporal graph neural network according to claim 2 is characterized in that: The method of obtaining low-resolution meteorological grid data, high-resolution land static grid data and high-resolution meteorological grid data according to the divided grid, the low-resolution meteorological data and the high-resolution land static data includes: Interpolating the low-resolution meteorological data to the divided grids to obtain low-resolution meteorological grid data; Calculating high-resolution land static grid data according to the divided grids and the high-resolution land static data; The high-resolution land static grid data and the low-resolution meteorological grid data are input into a preset meteorological research and forecasting modeling system to obtain high-resolution meteorological grid data.
4. The meteorological data downscaling method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The step of constructing a downscaling dataset based on the meteorological observation dataset, the gridded dataset and corresponding characteristic variables includes: According to the determined number of lag steps, the time lag variable corresponding to the current time is obtained; According to the meteorological observation dataset, the gridded dataset, the corresponding characteristic variables and the corresponding time lag variables, a lag time characteristic dataset is constructed as a dataset for downscaling.
5. The meteorological data downscaling method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The first spatiotemporal graph neural network includes an embedding layer, a graph structure building module, a spatiotemporal convolution module, an attention module and an output layer. The embedding layer of the first spatiotemporal graph neural network has the functions of spatiotemporal coding, average coding and feature embedding, and the output layer of the first spatiotemporal graph neural network includes a two-dimensional convolutional layer and a fully connected layer in the time and space dimensions.
6. The meteorological data downscaling method based on spatiotemporal graph neural network according to claim 1 is characterized in that: After inputting the data set for downscaling into the trained first spatiotemporal graph neural network, and downscaling the low-resolution meteorological grid data using the first spatiotemporal graph neural network to obtain predicted high-resolution meteorological grid data, the method further includes: The data set used for downscaling and the predicted high-resolution meteorological grid data are input into a trained second spatiotemporal graph neural network, and the predicted high-resolution meteorological grid data is corrected for deviation using the second spatiotemporal graph neural network to obtain corrected high-resolution meteorological grid data.
7. The meteorological data downscaling method based on spatiotemporal graph neural network according to claim 6 is characterized in that: The second spatiotemporal graph neural network includes a data processing module, a spatiotemporal representation learning module, an uncertainty modeling module and an extrapolation generation module.
8. A meteorological data downscaling device based on spatiotemporal graph neural network, characterized in that: include: A data acquisition unit, used to acquire meteorological observation data sets and grid data sets of meteorological stations in the study area, wherein the grid data sets include low-resolution meteorological grid data, high-resolution meteorological grid data and high-resolution land static grid data; A data preprocessing unit, used to perform data preprocessing on the meteorological observation data set and the gridded data set, and determine corresponding characteristic variables using a preset correlation analysis algorithm; A data set construction unit, used to construct a data set for downscaling according to the meteorological observation data set, the gridded data set and corresponding characteristic variables; A data downscaling unit is used to input the data set for downscaling into a trained first spatiotemporal graph neural network, and use the first spatiotemporal graph neural network to downscale the low-resolution meteorological grid data to obtain predicted high-resolution meteorological grid data.
9. An electronic device, characterized in that: It includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the meteorological data downscaling method based on a spatiotemporal graph neural network according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the meteorological data downscaling method based on a spatiotemporal graph neural network as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Aviation big data preprocessing technology
CN112766300A
Sub-season high-temperature weather forecast correction method based on cascade residual convolutional neural network
CN113344289A
PM2.5 concentration space-time change prediction method and system based on space-time diagram neural network
CN113919231A
Method for forecasting daily maximum temperature based on full-connection neural network
CN115526413A
Weather forecast downscaling algorithm
CN116151015A
Cited By
Precipitation space downscaling method based on geographic prior embedding
CN119537850A
Surface temperature refined downscaling method combined with space-time hysteresis
CN119691696A
A refined downscaling method of land surface temperature combining temporal and spatial lags
CN119691696B
Laryngoscope image super-resolution improving method, system and equipment based on deep learning
CN119693371A
Wind field downscaling simulation prediction system and method based on machine learning
CN119808541A
