Climate data correction method and system

Abstract

The invention relates to the technical field of climate data processing, and discloses a climate data correction method and system, and the method comprises the steps: obtaining first to-be-corrected data; performing a first screening operation on the first to-be-corrected data to obtain a first screening result, and generating three-channel feature data; inputting the three-channel feature data into a pre-trained first correction network model to obtain three-channel feature data correction output; performing a first division extraction operation on the first to-be-corrected data to obtain a regional feature matrix; and performing first fusion operation on the three-channel feature data correction output and the regional feature matrix to obtain multi-source climate data collaborative correction output. According to the method, the climate data correction precision can be remarkably improved, correction errors are reduced, high-reliability data support is provided for climate prediction, disaster early warning, agricultural production and new energy development, and the development of meteorological services towards the precision and intelligence direction is powerfully promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of climate data processing, and in particular to a climate data correction method and system. Background Art

[0002] Currently, meteorological forecasting and resource assessment methods face significant challenges when processing high-resolution, multi-dimensional climate data. While existing numerical simulation and statistical methods can provide basic meteorological forecasting capabilities, they have significant limitations in interpreting complex terrain and processing high-resolution gridded data. Furthermore, they lack the accuracy to model long-term spatiotemporal evolution and assess resource availability. These issues have become core bottlenecks hindering new energy development and power grid planning in Guizhou.

[0003] Traditional data correction methods mostly rely on complex physical models or statistical methods, and have the following defects: First, they have strict requirements on the quality and quantity of input data, and are difficult to adapt to the spatiotemporal dynamic changes of meteorological scenes; second, the high computational complexity leads to insufficient real-time performance, and the lack of regional feature modeling capabilities cannot meet the customized correction needs of multiple geographical regions; third, the multi-source climate data fusion mechanism is imperfect, resulting in limited accuracy and reliability of the correction results.

[0004] Therefore, there is an urgent need to develop an efficient, accurate, and adaptive climate data correction method that meets the following technical requirements: ① Through in-depth regional feature mining and adaptive modeling, customized corrections can be achieved for different geographic climate characteristics, improving data quality and the ability to represent spatial heterogeneity; ② Developing a dynamic fusion mechanism for multi-source data to enhance the spatiotemporal consistency and stability of the correction results; and ③ Optimizing the algorithm architecture to reduce computational complexity and meet real-time business processing requirements. The implementation of this method will overcome the technical barriers to new energy resource assessment in Guizhou, significantly improve the accuracy and practicality of climate data corrections, and provide reliable data support for optimizing climate prediction models, upgrading disaster warning systems, and assessing agricultural productivity, thereby promoting the evolution of meteorological services towards precision and intelligence. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a climate data correction method and system to solve the problem that existing climate data correction methods have low correction accuracy, high model complexity, and strong dependence on data.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a climate data correction method, comprising:

[0009] Obtaining first data to be corrected;

[0010] Performing a first screening operation on the first data to be corrected to obtain a first screening result and generate three-channel feature data;

[0011] Inputting the three-channel feature data into a pre-trained first correction network model to obtain a three-channel feature data correction output;

[0012] performing a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix;

[0013] A first fusion operation is performed on the three-channel feature data correction output and the regional feature matrix to obtain a multi-source climate data collaborative correction output.

[0014] As a preferred embodiment of the climate data correction method of the present invention, the first screening operation includes:

[0015] Processing the first data to be corrected using a first analysis method to obtain a first processing result;

[0016] The first processing result is input into a first regression model for screening to obtain a first screening result.

[0017] As a preferred embodiment of the climate data correction method of the present invention, the generating of three-channel characteristic data includes:

[0018] Based on the prediction results of the first regression model, the top two meteorological factors that have the greatest impact on the prediction results are determined, and each climate variable and its meteorological factor are constructed as three-channel feature data.

[0019] As a preferred solution of the climate data correction method described in the present invention, the predicted data and the three-channel characteristic data are pre-processed by standardization and then combined into a multi-channel meteorological variable matrix as the input data of the pre-trained first correction network model.

[0020] As a preferred embodiment of the climate data correction method of the present invention, the step of obtaining the corrected output of the three-channel characteristic data includes:

[0021] Inputting the multi-channel meteorological variable matrix into the spatiotemporal feature extraction module of the first layer to extract the first-level spatiotemporal features; performing a downsampling operation on the first-level spatiotemporal features, converting the high-resolution feature matrix into a low-resolution feature matrix using a subgrid segmentation method, and splicing them according to the channel dimension;

[0022] Inputting the spliced ​​feature matrix into the spatiotemporal feature extraction module of the second layer to extract secondary spatiotemporal features; performing a downsampling operation on the secondary spatiotemporal features, converting the high-resolution feature matrix into a low-resolution feature matrix using a subgrid segmentation method, and performing an upsampling operation after splicing according to the channel dimension to obtain a secondary upsampled feature matrix;

[0023] After splicing the secondary upsampling feature matrix and the secondary spatiotemporal features according to the channel dimension, a secondary upsampling fusion feature matrix is ​​obtained by a feature fusion operation; an upsampling operation is performed on the secondary upsampling fusion feature matrix to obtain a primary upsampling feature matrix;

[0024] The first-level up-sampled feature matrices are downscaled respectively and then spliced ​​according to the channel dimension. A final downscaling process is performed, and a sub-grid segmentation method is used to generate a three-channel feature data correction output.

[0025] As a preferred embodiment of the climate data correction method of the present invention, the first division and extraction operation includes:

[0026] A first clustering algorithm is used to perform unsupervised cluster analysis on the first data to be corrected and the label data to generate a region division result.

[0027] As a preferred embodiment of the climate data correction method of the present invention, the first division and extraction operation further comprises:

[0028] According to the regional division results, the difference of statistical characteristics in each region is calculated as a regional characteristic constraint item;

[0029] Constructing a convolutional neural network model, introducing the regional feature constraint term into the loss function, and optimizing the regional feature capture capability of the model by minimizing the regional feature difference between the first to-be-corrected data and the labeled data;

[0030] The shared regional features in the data are extracted through multi-layer convolution operations, and the shared regional features are output as a regional feature matrix.

[0031] In a second aspect, the present invention provides a climate data correction system, comprising:

[0032] A data acquisition unit, configured to acquire first data to be corrected;

[0033] a screening processing unit, configured to perform a first screening operation on the first data to be corrected, obtain a first screening result, and generate three-channel feature data; input the three-channel feature data into a pre-trained first correction network model to obtain a corrected output of the three-channel feature data;

[0034] a partitioning and extraction unit, configured to perform a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix;

[0035] The fusion output unit is used to perform a first fusion operation on the three-channel feature data correction output and the regional feature matrix to obtain a multi-source climate data collaborative correction output.

[0036] In a third aspect, the present invention provides a computer device comprising a memory and a processor; the memory is used to store computer-executable instructions, and the processor implements the steps of a climate data correction method when executing the computer-executable instructions.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a climate data correction method.

[0038] Compared with the existing technology, the beneficial effects of the present invention are as follows: the present invention provides a climate data correction method and system, which effectively simplifies the model structure, reduces the number of parameters, reduces the dependence on data, and enhances the adaptability and universality of the model in unknown data and complex meteorological scenarios by integrating the output of the RegCM4.7 climate model and the ERA5 reanalysis data, and introducing the Swin Transformer module and neural network architecture. At the same time, the method of the present invention can significantly improve the accuracy of climate data correction and reduce correction errors, providing highly reliable data support for climate forecasting, disaster warning, agricultural production and new energy development, and effectively promoting the development of meteorological services towards precision and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 This is a logical diagram of the overall process of the climate data correction method according to one embodiment of the present invention.

[0041] Figure 2 This is a general process framework diagram of a climate data correction method according to an embodiment of the present invention.

[0042] Figure 3 Schematic diagram of a multi-scale neural correction network of a climate data correction method according to an embodiment of the present invention.

[0043] Figure 4Schematic diagram of a downsampling module of a climate data correction method according to an embodiment of the present invention.

[0044] Figure 5 This is a graph of average wind speeds for the first 300 hours of 2013 using ERA5 data according to a climate data correction method according to an embodiment of the present invention.

[0045] Figure 6 This is a graph of average wind speeds for the first 300 hours of model output data for the climate data correction method according to one embodiment of the present invention.

[0046] Figure 7 Schematic diagram of a comparative experiment of a climate data correction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0048] Example 1, with reference to Figures 1-4 As an embodiment of the present invention, a climate data correction method is provided to solve the problems of low correction accuracy, high model complexity and strong dependence on data in existing methods. Figure 1 The specific steps shown include:

[0049] S100: Acquire first data to be corrected;

[0050] S200: performing a first screening operation on the first data to be corrected to obtain a first screening result and generate three-channel feature data;

[0051] S300: Inputting the three-channel feature data into a pre-trained first correction network model to obtain a correction output of the three-channel feature data;

[0052] S400: performing a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix;

[0053] S500: performing a first fusion operation on the three-channel feature data correction output and the regional feature matrix to obtain a multi-source climate data collaborative correction output.

[0054] It should be noted that the present invention provides a climate data correction method and system. By integrating the RegCM4.7 climate model output and ERA5 reanalysis data and introducing the Swin Transformer module and neural network architecture, it effectively simplifies the model structure, reduces the number of parameters, reduces the dependence on data, and enhances the model's adaptability and universality in unknown data and complex meteorological scenarios. At the same time, the method of the present invention can significantly improve the accuracy of climate data correction and reduce correction errors, providing highly reliable data support for climate forecasting, disaster warning, agricultural production, and new energy development, and effectively promoting the development of meteorological services towards precision and intelligence.

[0055] like Figure 2 The following is the overall process framework of the climate data correction method. Figure 2 A specific implementation of a climate data correction method is introduced in detail to illustrate the technical solution of the present invention.

[0056] S100: Acquire first data to be corrected;

[0057] In the embodiment of the present application, ERA5 is obtained as label data, CMIP6 data is obtained as background field data of the regional climate model RegCM4.7, and the RegCM4.7 regional climate model is run to obtain the climate model output data to be corrected, i.e., the first data to be corrected;

[0058] Specifically, CMIP6 data were obtained as background field data for the regional climate model RegCM4.7, including zonal wind speed (U wind component), meridional wind speed (V wind component), specific humidity, geopotential height, and air temperature. Sensitivity experiments were designed for specific geographical regions, and suitable parameterization schemes were selected to optimize simulation accuracy. Subsequently, the RegCM4.7 regional climate model was run to obtain climate model output data for correction. The generated simulation data covers the time span from 2006 to 2030, with a temporal resolution of 1 hour, a spatial resolution of 20 kilometers, a grid number of 40×40, and a coverage area of ​​800 km × 800 km rectangular range centered at 27°E, 106.5°N.

[0059] Specifically, ERA5 reanalysis data for labeling was obtained as the benchmark data for the calibration model; and the obtained multi-source climate data and ERA5 data were preprocessed to align the temporal and spatial resolutions to a 40×40 grid resolution to ensure data consistency.

[0060] It should be noted that the above step S100 ensures the comprehensiveness and timeliness of the source data, provides a basic guarantee for subsequent data screening, feature extraction and correction processing, and helps to improve the accuracy and reliability of the final climate data correction.

[0061] S200: performing a first screening operation on the first data to be corrected to obtain a first screening result and generate three-channel feature data;

[0062] In an optional embodiment, the first filtering operation may be time window filtering, filtering the first data to be corrected according to the time dimension, such as selecting data for a specific time period, or extracting high-frequency time series data, which can highlight climate characteristics within a specific time range. The first filtering operation may also be spatial region filtering, filtering the data according to a geographic spatial range, such as retaining only meteorological data for the Guizhou region, or further refining the data to a specific altitude layer, latitude and longitude grid, or terrain feature area, which helps to focus on the climate characteristics of the target area.

[0063] In an optional embodiment, the first screening operation can also be variable type screening, which is performed according to the type of climate variables. For example, variables closely related to wind energy and solar energy are extracted from the first data to be corrected, and these variables are respectively assigned to different channels to form three-channel characteristic data, which can specifically capture the changing characteristics of key climate elements.

[0064] In an embodiment of the present application, the steps of the first screening operation include: using a first analysis method to process the first data to be corrected to obtain a first processing result; inputting the first processing result into a first regression model for screening to obtain a first screening result.

[0065] In an optional embodiment, the first analysis method may be a principal component analysis method, a wavelet transform method, or an empirical mode decomposition method.

[0066] In an optional embodiment, the first regression model can be a linear regression model, which performs simple linear screening on the features in the first processing result, identifies which variables have a significant impact on the target output, and eliminates irrelevant or redundant features; the first regression model can also be a ridge regression model. When there are highly correlated features in the first processing result, ridge regression can effectively reduce model complexity and improve stability, thereby achieving feature screening and weighing; the first regression model can also be a random forest regression model, which can screen out the features that have the greatest impact on the prediction results by calculating the importance score of each feature, while ignoring noise or irrelevant variables.

[0067] In the embodiment of the present application, the first analysis method is the spatiotemporal principal component analysis (ST-PCA) method. The spatiotemporal principal component analysis method is used to reduce the dimensionality of the input data, process the data of each variable at each time point, obtain a principal component score and construct a principal component score sequence to capture the key spatiotemporal features in the data, effectively achieve efficient feature decoupling and information compression of multidimensional spatiotemporal data, significantly reduce computational complexity and suppress noise interference;

[0068] In the embodiment of the present application, the first regression model is a linear regression model. The principal component score sequence of the data to be predicted is input into the linear regression model, the principal component score sequence of the data to be predicted is used as the dependent variable, and the principal component score sequences of the remaining climate variables are input one by one as independent variables. A linear regression model is constructed and key meteorological factors are screened. Based on the comparison results of the determination coefficients, the top two meteorological factors with the greatest impact on the prediction results are determined, and each climate variable and its meteorological factor are combined into three-channel feature data as the core input features of the subsequent prediction model. The relevant calculations include:

[0069]

[0070] Among them, F represents the meteorological factor, Proj ST-PCA (X) represents spatiotemporal principal component analysis, represents the data alignment operation, W is the regression coefficient matrix;

[0071] In the embodiment of the present application, a spatiotemporal principal component analysis method is used to extract principal component scores of different variables. The specific steps include:

[0072] Preprocess the input data. In meteorological factor screening, data centering is not only used to simplify mathematical operations, but also to eliminate dimensional differences and systematic biases between different meteorological factors. For example, the numerical ranges of temperature and precipitation vary greatly. After centering, the contribution of each factor to the target variable can be more fairly assessed. The formula is expressed as:

[0073]

[0074] Among them, X centered is the centralized data, X is the climate model output data to be corrected, is the mean matrix of the data;

[0075] Calculate the covariance matrix. The covariance matrix is ​​a key tool for screening meteorological factors. Its off-diagonal elements directly reflect the linear correlation between factors. For example, the covariance between temperature and sea level pressure is negative, indicating that the two are inversely correlated. This is consistent with the thermodynamic coupling mechanism in meteorology. The formula is expressed as:

[0076]

[0077] Among them, C is the covariance matrix, T is the length of the time dimension, Transpose of centralized data;

[0078] The covariance matrix is ​​decomposed into eigenvectors and eigenvalues ​​to reveal the intrinsic structure of the data. The size of the eigenvalue reflects the importance of the principal component represented by the corresponding eigenvector, while the eigenvector indicates the distribution direction of the data in the direction of the principal component. For example, in precipitation prediction, a certain eigenvector has a large and positive load on the humidity factor, indicating that high humidity is positively correlated with heavy precipitation. The formula is expressed as:

[0079] C=QΛQ T

[0080] Where Q is the eigenvector matrix and Λ is the diagonal matrix;

[0081] The principal components are extracted by selecting eigenvectors with larger eigenvalues; the purpose is to project the original data into the low-dimensional space defined by the eigenvectors, thereby achieving data dimensionality reduction. By selecting the main eigenvectors, the most important information in the data can be retained while removing noise and redundant information, improving the efficiency of data processing and the accuracy of analysis. The formula is expressed as:

[0082] Y=x centered Q

[0083] Among them, Y is the principal component matrix after dimensionality reduction, and each column of Y k Represents the score of the data on the kth principal component.

[0084] In the embodiment of the present application, a linear regression model is used to evaluate the two meteorological factors that have the greatest impact on the predicted variables. The specific steps include:

[0085] Assuming a linear relationship between the dependent variable Y and the independent variable, a quantitative relationship between the dependent and independent variables is established by estimating the regression coefficient. The magnitude and sign of the regression coefficient reflect the direction and strength of the contribution of meteorological factors. For example, if the dependent variable is precipitation, the regression coefficient β1 of the independent variable humidity is positive and large, indicating that humidity has a significant positive contribution to precipitation; while the regression coefficient β2 of the independent variable vertical wind speed is negative, indicating that vertical wind speed has an inhibitory effect. The formula is as follows:

[0086] Y=β0+β1X1+β2X2+…+β p X p +∈

[0087] Among them, Y is the dependent variable, which represents the target variable of the meteorological factor to be extracted, β0 is the intercept term of the model, β1, β2, ..., β p are the regression coefficients of the corresponding meteorological elements, reflecting the linear contribution of each meteorological factor to the dependent variable. ∈ is the random error term, which obeys ∈~N(0,σ 2 ).

[0088] By minimizing the residual sum of squares to solve the regression coefficient, in the meteorological factor screening, minimizing RSS is not only to improve the model fitting accuracy, but also to identify meteorological factors that have a significant contribution to the target variable (such as precipitation, temperature). Through the change of RSS, the predictive ability of different factor combinations can be evaluated, thereby screening out key factors. The formula is expressed as:

[0089]

[0090] Where n is the sample size, is the model's predicted value for the i-th sample;

[0091] By using the least squares method, the analytical solution of the regression coefficient is derived:

[0092]

[0093] Where X is the data design matrix (each column of the design matrix corresponds to a candidate meteorological factor (such as temperature, humidity, wind speed, etc.), and each row corresponds to an observation value at a time point or space point), and Y is the dependent variable vector.

[0094] Goodness of fit (R 2 ), the goodness of fit indicates the proportion of the variation explained by the model to the total variation. The closer its value is to 1, the stronger the model’s ability to explain the data. For example, in precipitation prediction, if the R of a factor combination is 2 Close to 0.8, indicating that these factors can explain 80% of precipitation variation and have high predictive ability. By calculating the goodness of fit, the goodness of fit of the model can be quantitatively evaluated. The formula is as follows:

[0095]

[0096] Among them, TSS is the total sum of squares, R 2 The closer it is to 1, the stronger the explanatory power of the model;

[0097] The regression coefficient significance (t-test) method is used to evaluate the significant contribution of each meteorological factor to the target variable:

[0098]

[0099] If |t j |>t α / 2 (np-1), then reject H0: β j =0, indicating that the meteorological factor has a significant impact on the predicted variable. Through the t-test, we can screen out the independent variables that have a significant contribution to the model, thereby optimizing the model structure.

[0100] It should be noted that the above step S200 can accurately extract characteristic information related to climate data correction, reduce the interference of irrelevant data, and at the same time provide structured input for subsequent model processing by constructing three-channel characteristic data, effectively improving data processing efficiency and correction accuracy.

[0101] S300: Inputting the three-channel feature data into a pre-trained first correction network model to obtain a correction output of the three-channel feature data;

[0102] In an optional embodiment, the first correction network model can be a convolutional neural network model, which extracts local patterns and complex nonlinear relationships in the three-channel feature data through convolutional layers, thereby achieving accurate correction of climate data; the first correction network model can be a Swin Transformer, which can capture global dependencies and local details in the three-channel feature data, and gradually refine feature information through hierarchical design, which is suitable for high-precision climate data correction tasks; the first correction network model can also be a recurrent neural network. If the three-channel feature data has time series characteristics, a recurrent neural network model can be used to effectively model dynamic changes in the time series and capture long-term dependencies in historical data, thereby improving correction accuracy.

[0103] In an embodiment of the present application, the first correction network model integrates the Swin Transformer module and the neural network architecture to effectively capture the complex nonlinear characteristics in the climate data and improve the prediction accuracy through the correction learning mechanism.

[0104] Specifically, such as Figure 3 As shown, the processing steps of the first revised network model include:

[0105] After standardization preprocessing, the forecast data and three-channel feature data are combined into a multi-channel meteorological variable matrix (resolution of 3×40×40), which serves as the input data of the pre-trained first correction network model.

[0106] The multi-channel meteorological variable matrix is ​​input into the four ST (spatiotemporal) feature extraction modules in the first layer. Through spatiotemporal convolution and pooling operations, four first-level ST features (with a resolution of 3×40×40) are extracted to capture the spatiotemporal dependencies in the data. The ST feature extraction module can simultaneously consider the spatial and temporal dimensions of the data, capturing the spatial distribution changes and temporal evolution of meteorological variables through spatiotemporal convolution operations. Pooling operations help reduce feature dimensionality and enhance feature robustness.

[0107] The 4 first-level ST features are downsampled, and the high-resolution feature matrix is ​​converted into a low-resolution feature matrix using the sub-grid segmentation method to obtain 4 first-level feature matrices (resolution of 6×20×20) to reduce the computational complexity and avoid the waste of computing resources and long model training time caused by excessive data volume. The schematic diagram of the downsampling module is shown in Figure 4 As shown;

[0108] The four first-level feature matrices are concatenated according to the channel dimension and input into the two ST feature extraction modules in the second layer. Through further spatiotemporal convolution and pooling operations, two second-level ST features (resolution of 24×20×20) are extracted. Concatenating multiple first-level feature matrices according to the channel dimension can integrate the information from different first-level features to form a richer feature representation and enhance the ability to learn complex spatiotemporal patterns in the data.

[0109] The two secondary ST features are downsampled and the high-resolution feature matrix is ​​converted into a low-resolution feature matrix using the sub-grid segmentation method to obtain two secondary feature matrices (with a resolution of 48×10×10), which further reduces the computational burden.

[0110] The two secondary feature matrices are spliced ​​according to the channel dimension, and two independent upsampling operations are performed on each. The low-resolution feature matrix is ​​restored to a high-resolution feature matrix using the bilinear interpolation method to obtain two secondary upsampled feature matrices (resolution of 48×20×20). This ensures that the feature matrix maintains data integrity and accuracy while improving the resolution.

[0111] The two secondary upsampling feature matrices are concatenated with the two secondary ST features according to the channel dimension. Through the feature fusion operation, two secondary upsampling fusion feature matrices (resolution 72×20×20) are obtained. By integrating the feature information at different resolutions, multi-scale feature fusion is achieved, so that the model can simultaneously capture the changes and characteristics of meteorological data at different scales, and enhance the model's multi-dimensional understanding of complex meteorological phenomena.

[0112] Four independent upsampling operations are performed on the two secondary upsampling fusion feature matrices. The low-resolution feature matrix is ​​restored to a high-resolution feature matrix using bilinear interpolation. At the same time, it is spliced ​​according to the channel dimension to obtain a first-level upsampling feature matrix (resolution 96×40×40). Through multiple independent upsampling operations, feature information at different resolutions is integrated to achieve multi-scale feature fusion, and the resolution of the feature matrix is ​​gradually and finely restored to a higher resolution level.

[0113] The first-level upsampled feature matrix is ​​gradually downscaled, and the high-resolution feature matrix is ​​converted into a low-resolution feature matrix using the subgrid segmentation method. Four downscaled feature matrices (resolution of 3×40×40) are obtained to obtain a resolution that matches the model output requirements;

[0114] The four downscaled feature matrices are spliced ​​according to the channel dimension and downscaled. The subgrid segmentation method is used to generate the final corrected output (resolution of 1×40×40). Multiple downscaled feature matrices are spliced ​​according to the channel dimension and then downscaled. This can integrate all feature information and generate output data that meets application requirements.

[0115] In the embodiment of the present application, the specific composition of the first revised network model is:

[0116] Windowed Self-Attention: Meteorological data has significant spatial heterogeneity (such as temperature differences between mountainous areas and plains). Windowed self-attention can more accurately capture these local characteristics through local calculations. For example, in temperature correction in mountainous areas, windowed self-attention can identify the local impact of terrain on temperature. The formula is as follows:

[0117]

[0118] Among them, Q, K, and V are query (representing the target variable or its feature representation that needs to be corrected), key (other meteorological factors or spatial-temporal context information related to the target variable), and value (representing the features or corrections related to the target variable) matrices, respectively. k is the dimension of the key.

[0119] Shifted Window: Meteorological phenomena often span multiple windows. Shifted windows enhance the model's ability to perceive these large-scale features by sharing some features. By changing the starting position of the window, adjacent windows can share some features, thereby enhancing the model's ability to perceive cross-window features. The formula is as follows:

[0120]

[0121] Where: x is the input feature map, and Roll is the cyclic shift operation.

[0122] Multilayer Perceptron (MLP): An important component of the windowed self-attention module. Meteorological data contains complex nonlinear relationships (such as the interaction between temperature and humidity). MLP can effectively model these relationships through nonlinear activation functions (such as GELU), which is used to further process features and enhance the model's nonlinear expression capabilities. The formula is as follows:

[0123] MLP(x)=W2·GELU(W1·x+b1)+b2

[0124] Where: W1 is the weight matrix that maps the input features to the hidden layer, W2 is the hidden matrix that maps the hidden layer features to the output features, b1 and b2 are bias terms;

[0125] Residual Links and Normalization Formula: Meteorological data is typically high-dimensional and noisy. Residual links avoid the vanishing gradient problem in deep network training by preserving input features. Layer normalization further stabilizes the training process by normalizing each feature channel. This design not only improves model training efficiency but also enhances model robustness. The formula is as follows:

[0126] x out =LayerNorm(x+MLP(x))

[0127] Where x is the input feature of meteorological data.

[0128] It should be noted that the above step S300 uses the optimized model structure and algorithm to efficiently and accurately correct the climate data, reduce the number of parameters and dependence on data, enhance the adaptability and generalization ability of the model when dealing with complex meteorological scenarios, and thus significantly improve the accuracy and reliability of the correction results.

[0129] S400: performing a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix;

[0130] In this embodiment of the present application, the first partitioning and extraction operation includes:

[0131] Using the first clustering algorithm to perform unsupervised cluster analysis on the first data to be corrected and the labeled data, generating a regional division result to capture the spatial heterogeneity of the data and avoid bias introduced by artificial prior assumptions through an unsupervised learning mechanism;

[0132] Based on the regional division results, the difference in statistical characteristics within each region, including the mean and variance, is calculated as the regional characteristic constraint term to construct the subsequent optimization objective function. By quantitatively extracting regional statistical characteristics, a physically interpretable spatial constraint representation system is established.

[0133] Construct a convolutional neural network model and introduce a regional feature constraint term into the loss function. By minimizing the regional feature difference between the first to-be-corrected data and the labeled data, the model's regional feature capture capability is optimized.

[0134] The shared regional features in the data are extracted through multi-layer convolution operations, and the shared regional features are output as a regional feature matrix for subsequent data fusion and correction, ensuring the effective expression of regional features in the final correction results.

[0135] In an optional embodiment, the first clustering algorithm may be hierarchical clustering, which implements clustering by constructing a tree structure. It may be agglomerative or divisive, does not require the number of clusters to be determined in advance, and can provide more detailed information about the relationship between data, and is suitable for exploring the spatial heterogeneity of climate data at different scales. The first clustering algorithm may be K-Means clustering, which divides data into K clusters by minimizing the square distance between points within a cluster. It is suitable for finding groups of data points with similar characteristics and can effectively capture spatial distribution patterns in meteorological data. The first clustering algorithm may also be DBSCAN, which performs clustering based on the density of data points rather than a predefined number of clusters. It is particularly suitable for identifying regional features with high density and irregular shapes in meteorological data, and can also mark data points that are considered noise.

[0136] In the embodiment of the present application, the first clustering algorithm is K-Means clustering, and the specific steps include:

[0137] The meteorological data points are divided into the nearest cluster centers to form spatially continuous areas. By minimizing the Euclidean distance between the data points and the cluster centers, efficient data division is achieved, providing a basis for subsequent feature extraction and processing. The formula is as follows:

[0138] C k ={x i : ‖x i -μ k ‖ 2 ≤‖x i -μ j ‖ 2 for all j≠k}

[0139] Among them, C k is the kth cluster, μ k is the center of the kth cluster, x i is a meteorological data point;

[0140] Cluster updating is achieved by calculating the mean of all data points in each cluster. This process helps to gradually optimize the location of the cluster center, thereby improving the accuracy and stability of clustering. The formula is as follows:

[0141]

[0142] In the embodiment of the present application, a convolutional neural network is used to obtain data region features. The specific steps include:

[0143] The convolution kernel of a preset size is used to perform a sliding window dot product operation on the input data to extract the local spatial features of the input data layer by layer. The weight parameters of the convolution kernel are optimized and updated through the back-propagation algorithm to maximize the discriminative ability of feature extraction. By setting the convolution kernel size to 3×3 and the padding to 1, the spatial resolution of the feature map is ensured to remain consistent before and after convolution. At the same time, a three-layer cascade convolution architecture is adopted to realize cross-scale mapping of the number of feature channels from 1 to 16, 16 to 32, and 32 to 1. The ReLU activation function is applied after each convolution operation to introduce nonlinear transformation to enhance the feature expression ability. The formula is as follows:

[0144]

[0145] Among them, y[i,j] is the value of the i-th row and j-th column of the output feature map, x[i,j,k] is the value of the i-th row, j-th column and k-th channel of the input data, and w[m,n,k] is the value of the m-th row, n-th column and k-th channel of the convolution kernel. b is the bias term, K is the number of channels of the input meteorological data, and M and N are the sizes of the convolution kernel.

[0146] In the embodiment of the present application, a loss function with an added regional feature constraint term is used to obtain data regional features. The specific formula is as follows:

[0147]

[0148] Among them, y i is the label data, is the data to be corrected, λ is the constraint weight, K is the number of regions, μ y,k is the mean of the k regions of the label data, is the mean value of the k regions to be corrected, is the variance of the label data k region, The variance of the k region to be corrected.

[0149] It should be noted that the above step S400 can accurately extract the climate characteristics of a specific area, enhance the understanding and analysis of local meteorological conditions, and provide highly targeted and detailed regional characteristic information for subsequent data fusion, which helps to improve the accuracy and precision of the overall climate data correction.

[0150] S500: performing a first fusion operation on the three-channel feature data correction output and the regional feature matrix to obtain a multi-source climate data collaborative correction output.

[0151] In an embodiment of the present application, the arithmetic average method is used to fuse the three-channel feature data correction output and the regional feature matrix to obtain a collaborative correction output of multi-source climate data.

[0152] It should be noted that the above step S500 integrates global and local climate characteristic information to achieve effective combination and complementarity of multi-source data, thereby improving the comprehensiveness and accuracy of climate data correction and providing strong support for precise meteorological services.

[0153] Example 2, reference Figures 5 to 7 Based on the previous embodiment, this embodiment provides an application example of a climate data correction method and system to verify and illustrate the technical effects adopted in this method.

[0154] like Figure 5 and Figure 6 As shown in the figure, the data simulation for the 2013 period has a time resolution of 1 hour, a spatial resolution of 20km×20km, a number of attention heads of 3, and a batch size of 17. As can be seen from the figure, the wind speed during this period is mainly concentrated in the upper left and upper right corners, the wind speed in the middle area is higher, and the wind speed in the upper and lower areas is lower. The model output is relatively smooth in certain areas, and Figure 5 In comparison, the areas in the upper left and upper right corners are not prominent enough; however, the data better reflects the main spatial characteristics of the wind speed during this period, which is high in the upper left and upper right corners, and low above and below.

[0155] Figure 7 This figure shows the performance of different models at three different time points in 2013 (February, June, and October): four performance metrics: Euclidean distance, cosine similarity, mean squared error, and Pearson correlation coefficient. Each time point corresponds to four sub-graphs, one for each of these four performance metrics. By comparing the performance of different models using the same metrics, we can evaluate the performance of each model.

[0156] Performance indicator analysis in February:

[0157] Euclidean distance: This model performs best on the Euclidean distance metric, with the lowest distance value, indicating that the predictions are least different from the real data. In contrast, the original data has the highest Euclidean distance, indicating that its prediction error is the largest.

[0158] Cosine similarity: In terms of cosine similarity, both the proposed model and the CNN model achieved a high similarity, close to 1, indicating that the predictions of the two models are very close to the direction of the real data. The similarity of the original data is relatively low, indicating that the direction of the predictions is quite different from that of the real data.

[0159] Mean Square Error: This model has the lowest mean square error, indicating that the sum of squared errors in its prediction results is the smallest and the prediction accuracy is the highest. The original data has the highest mean square error and the lowest prediction accuracy.

[0160] Pearson correlation coefficient: The Pearson correlation coefficients for this model and the CNN model are high, indicating that the predictions of these two models have a high linear correlation with the real data. The Pearson correlation coefficient for the original data is negative, indicating that the predictions are negatively correlated with the real data.

[0161] Performance indicator analysis in June:

[0162] Euclidean distance: This model still performs best in the Euclidean distance metric, followed by the U-NET model, and the CNN model performs worst.

[0163] Cosine similarity: This model and the CNN model performed better in the cosine similarity index, the U-NET model performed slightly worse, and the original data performed the worst;

[0164] Mean square error: This model has the lowest mean square error, followed by the U-NET model. The mean square error between the CNN model and the original data is higher.

[0165] Pearson Correlation Coefficient: This model has the highest Pearson Correlation Coefficient, indicating that its predictions have the strongest linear correlation with the real data. The U-NET model performs second, the CNN model performs average, and the original data performs the worst.

[0166] Performance indicator analysis in October:

[0167] Euclidean distance: This model performs best in the Euclidean distance metric, followed by the U-NET model, and the CNN model performs worst.

[0168] Cosine similarity: This model and the CNN model performed better in the cosine similarity index, the U-NET model performed slightly worse, and the original data performed the worst;

[0169] Mean square error: This model has the lowest mean square error, followed by the U-NET model. The mean square error between the CNN model and the original data is higher.

[0170] Pearson Correlation Coefficient: This model has the highest Pearson Correlation Coefficient, indicating that its predictions have the strongest linear correlation with the real data. The U-NET model performs second, the CNN model performs average, and the original data performs the worst.

[0171] From the above analysis, it can be seen that by integrating the RegCM4.7 climate model output and ERA5 reanalysis data and introducing the Swin Transformer module and neural network architecture, the present invention effectively simplifies the model structure, reduces the number of parameters, reduces the dependence on data, and enhances the model's adaptability and universality in unknown data and complex meteorological scenarios. At the same time, the method of the present invention can significantly improve the accuracy of climate data correction and reduce correction errors, providing highly reliable data support for climate forecasting, disaster warning, agricultural production, and new energy development, and effectively promoting the development of meteorological services towards precision and intelligence.

[0172] Example 3: This example provides a climate data correction system, including:

[0173] A data acquisition unit, configured to acquire first data to be corrected;

[0174] a screening processing unit, configured to perform a first screening operation on the first data to be corrected, obtain a first screening result, and generate three-channel feature data; input the three-channel feature data into a pre-trained first correction network model to obtain a corrected output of the three-channel feature data;

[0175] a partitioning and extraction unit, configured to perform a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix;

[0176] The fusion output unit is used to perform a first fusion operation on the three-channel feature data correction output and the regional feature matrix to obtain the multi-source climate data collaborative correction output.

[0177] It should be noted that the technical solution of the climate data correction system and the technical solution of the above-mentioned climate data correction method belong to the same concept. For details not described in detail in the technical solution of the climate data correction system in this embodiment, please refer to the description of the technical solution of the above-mentioned climate data correction method.

[0178] The above-mentioned unit modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of the above-mentioned modules.

[0179] This embodiment also provides an electronic device, which includes a processor, memory, a communication interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (near field communication), or other technologies. When the computer program is executed by the processor, a method for correcting climate data is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a key, trackball, or touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse.

[0180] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0181] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0182] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the method of the embodiment of the present invention.

[0183] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0184] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages.

[0185] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0186] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0187] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0188] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0189] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A climate data correction method, characterized in that: include: Obtaining first data to be corrected; Performing a first screening operation on the first data to be corrected to obtain a first screening result and generate three-channel feature data; Inputting the three-channel feature data into a pre-trained first correction network model to obtain a three-channel feature data correction output; performing a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix; A first fusion operation is performed on the three-channel feature data correction output and the regional feature matrix to obtain a multi-source climate data collaborative correction output.

2. A climate data correction method according to claim 1, characterized in that: The first screening operation includes: Processing the first data to be corrected using a first analysis method to obtain a first processing result; The first processing result is input into a first regression model for screening to obtain a first screening result.

3. A climate data correction method according to claim 1, characterized in that: Generating three-channel feature data includes: Based on the prediction results of the first regression model, the top two meteorological factors that have the greatest impact on the prediction results are determined, and each climate variable and its meteorological factor are constructed as three-channel feature data.

4. A climate data correction method according to claim 3, characterized in that: After the predicted data and the three-channel characteristic data are subjected to standardization preprocessing, they are combined into a multi-channel meteorological variable matrix as input data of the pre-trained first correction network model.

5. A climate data correction method according to claim 4, characterized in that: The obtaining of the three-channel feature data correction output includes: Inputting the multi-channel meteorological variable matrix into the spatiotemporal feature extraction module of the first layer to extract the first-level spatiotemporal features; performing a downsampling operation on the first-level spatiotemporal features, converting the high-resolution feature matrix into a low-resolution feature matrix using a subgrid segmentation method, and splicing them according to the channel dimension; Inputting the spliced ​​feature matrix into the spatiotemporal feature extraction module of the second layer to extract secondary spatiotemporal features; performing a downsampling operation on the secondary spatiotemporal features, converting the high-resolution feature matrix into a low-resolution feature matrix using a subgrid segmentation method, and performing an upsampling operation after splicing according to the channel dimension to obtain a secondary upsampled feature matrix; After splicing the secondary upsampling feature matrix and the secondary spatiotemporal features according to the channel dimension, a secondary upsampling fusion feature matrix is ​​obtained by a feature fusion operation; an upsampling operation is performed on the secondary upsampling fusion feature matrix to obtain a primary upsampling feature matrix; The first-level up-sampled feature matrices are downscaled respectively and then spliced ​​according to the channel dimension. A final downscaling process is performed, and a sub-grid segmentation method is used to generate a three-channel feature data correction output.

6. A climate data correction method according to claim 1, characterized in that: The first partitioning and extracting operation includes: A first clustering algorithm is used to perform unsupervised cluster analysis on the first data to be corrected and the label data to generate a region division result.

7. A climate data correction method according to claim 6, characterized in that: The first partitioning and extracting operation further includes: According to the regional division results, the difference of statistical characteristics in each region is calculated as a regional characteristic constraint item; Constructing a convolutional neural network model, introducing the regional feature constraint term into the loss function, and optimizing the regional feature capture capability of the model by minimizing the regional feature difference between the first to-be-corrected data and the labeled data; The shared regional features in the data are extracted through multi-layer convolution operations, and the shared regional features are output as a regional feature matrix.

8. A climate data correction system, applying the method according to any one of claims 1 to 7, characterized in that: include: A data acquisition unit, configured to acquire first data to be corrected; a screening processing unit, configured to perform a first screening operation on the first data to be corrected, obtain a first screening result, and generate three-channel feature data; Inputting the three-channel feature data into a pre-trained first correction network model to obtain a three-channel feature data correction output; a partitioning and extraction unit, configured to perform a first partitioning and extraction operation on the first data to be corrected to obtain a regional feature matrix; The fusion output unit is used to perform a first fusion operation on the three-channel feature data correction output and the regional feature matrix to obtain a multi-source climate data collaborative correction output.

9. A computer device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor implements the steps of a climate data correction method according to any one of claims 1 to 7 when executing the computer-executable instructions.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer executable instructions are executed by a processor, the steps of a climate data correction method according to any one of claims 1 to 7 are implemented.

Images