Severe convective weather forecast result correction method and system

Abstract

The embodiment of the invention provides a severe convection weather forecast result correction method and system. The method is applied to the technical field of weather observation data processing, and comprises the following steps: a random forest regression model learns a mapping relationship between ERA5 grid data and meteorological station data to generate virtual meteorological grid data with similar precision to the meteorological station data; a dynamic space-time attenuation network is trained, and learning of space-time features is realized through 3D convolution, a space-time attention mechanism and a GRU module; through inputting virtual meteorological grid data and real-time site data, the network learns the time-space attenuation rate of severe convective weather, and establishes the relationship between the time-space characteristics and the forecast result. And applying the learned space-time attenuation rate to the severe convection prediction data, and correcting the result. In this way, the technical problem of how to effectively utilize real-time observation data to correct severe convection weather forecast in the prior art can be solved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of weather observation data processing, and in particular to a method and system for correcting severe convective weather forecast results. Background Art

[0002] At present, the forecast of severe convective weather (such as strong winds, heavy rainfall, etc.) has always been a core area of meteorological research. Existing numerical weather forecast models usually rely on meteorological reanalysis data such as ERA5. However, due to the low spatial resolution of these data, it is often difficult to meet the accuracy requirements of long-term series forecasts. Although meteorological station data have high temporal resolution and accuracy, their spatial distribution is uneven and coverage is insufficient. In addition, the method for using real-time data to correct forecast results is still unclear, which makes the application of real-time station data in severe convective weather prediction face challenges. Therefore, how to effectively use real-time observation data to correct severe convective weather forecasts has become a key issue in current research. Summary of the Invention

[0003] The present disclosure provides a method and system for correcting severe convective weather forecast results, which aims to solve the technical problem in the prior art of how to effectively use real-time observation data to correct severe convective weather forecasts.

[0004] According to a first aspect of the present disclosure, a method for correcting severe convective weather forecast results is provided, comprising:

[0005] Collect meteorological station data, ERA5 reanalysis meteorological data, and auxiliary data such as land cover type and elevation; preprocess meteorological station data, meteorological data, and auxiliary data; preprocess missing value interpolation and standardization;

[0006] The random forest regression model learns the mapping relationship between ERA5 grid data and meteorological station data, generating virtual meteorological grid data with similar accuracy to meteorological station data;

[0007] A dynamic spatiotemporal decay network is trained. The network learns spatiotemporal features through 3D convolution, spatiotemporal attention mechanism, and GRU module. By inputting virtual meteorological grid data and real-time station data, the network learns the spatiotemporal decay rate of severe convective weather and establishes the relationship between spatiotemporal features and forecast results.

[0008] The learned spatiotemporal attenuation rate is applied to the severe convection prediction data, and the spatiotemporal attenuation rate is used to correct the severe convective weather forecast results output by the severe convection numerical prediction model, and the results are corrected to output accurate severe convective weather forecast results.

[0009] According to a second aspect of the present disclosure, a severe convective weather forecast result correction system is provided, comprising:

[0010] Data acquisition module, used to collect meteorological station data, ERA5 reanalysis meteorological data, surface cover type, elevation auxiliary data; pre-process meteorological station data, meteorological data and auxiliary data;

[0011] Random forest regression module, used to convert ERA5 grid data into virtual meteorological grid data using random forest regression;

[0012] The dynamic spatiotemporal decay network module is used to train a dynamic spatiotemporal decay network. The dynamic spatiotemporal decay network learns spatiotemporal features through 3D convolution, spatiotemporal attention mechanism, and GRU module. By inputting virtual meteorological grid data and real-time station data, the dynamic spatiotemporal decay network learns the spatiotemporal decay rate of severe convective weather and establishes the relationship between spatiotemporal features and forecast results.

[0013] The result output module is used to apply the learned spatiotemporal attenuation rate to the severe convection prediction data, use the spatiotemporal attenuation rate to correct the severe convective weather forecast results output by the severe convection numerical prediction model, revise the results, and output accurate severe convective weather forecast results.

[0014] Compared with the prior art, the advantages and positive effects achieved by the present disclosure are:

[0015] The present disclosure utilizes real-time meteorological station data and ERA5 reanalysis data to achieve effective correction of severe convective forecasts through an efficient deep learning network. The core of this method is to use random forest regression to generate virtual meteorological grid data, and to learn and extract the spatiotemporal characteristics and attenuation rates of severe convective weather through the DSTDN network, thereby improving the forecast results. The present disclosure can be applied to the real-time correction of severe convective weather forecasts. For example, in the forecast of a heavy rainfall event, virtual meteorological grid data is generated through meteorological station and ERA5 data, and the spatiotemporal attenuation rates learned by the trained dynamic spatiotemporal attenuation network (DSTDN) are used to correct the original forecast results. The final output result can better reflect the actual weather conditions and provide the meteorological department with more accurate early warning information. In this way, the accuracy of severe convective weather forecasts is significantly improved, which is suitable for the meteorological department to quickly obtain high-precision weather information in a short time.

[0016] This paper combines real-time meteorological station data with ERA5 reanalysis data, and uses a random forest regression model to convert ERA5 grid data into virtual meteorological grid data with similar accuracy to meteorological station data. Subsequently, the dynamic spatiotemporal decay network (DSTDNet) is used based on the virtual meteorological grid data to obtain the spatiotemporal decay rate information of severe convection-related meteorological elements. Finally, the learned meteorological element decay rate is applied to the severe convection prediction data to achieve effective correction of severe convective weather forecast results. This method shows significant application potential in improving the accuracy of severe convective weather forecasts.

[0017] It should be understood that the contents described in the Summary of the Invention section are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0019] Figure 1 A flowchart of a method for correcting severe convective weather forecast results according to an embodiment of the present disclosure is shown;

[0020] Figure 2 A schematic diagram illustrating a method for correcting severe convective weather forecast results according to an embodiment of the present disclosure is shown;

[0021] Figure 3 A schematic diagram of constructing a random forest regression model according to an embodiment of the present disclosure is shown;

[0022] Figure 4 A schematic diagram of the structure of a dynamic spatiotemporal decay network (DSTDNet) according to an embodiment of the present disclosure is shown;

[0023] Figure 5 A block diagram of a severe convective weather forecast result correction system according to an embodiment of the present disclosure is shown;

[0024] Figure 6 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0025] To make the purpose, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.

[0026] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0027] Figure 1 FIG. 1 is a flow chart of a method 100 for correcting severe convective weather forecast results according to an embodiment of the present disclosure. Figure 1 As shown, the method 100 includes:

[0028] S110: Collect meteorological station data, ERA5 reanalyze meteorological data and auxiliary data such as surface cover type and elevation; preprocess meteorological station data, meteorological data and auxiliary data; preprocess missing value interpolation and standardization, etc.

[0029] Optionally, in some embodiments, the weather station data includes wind speed and precipitation, etc., and the weather data is analyzed to change the temperature, air pressure, humidity, wind and cloud cover, etc.

[0030] It should be noted that, in the embodiment, the process of pre-processing the weather station data, weather data and auxiliary data specifically includes the following steps:

[0031] First, raw data are collected from meteorological stations, ERA5 reanalysis data, and auxiliary data sources such as land cover types and elevation. These data may contain information of different time scales, spatial resolutions, and formats. The purpose of the initial screening is to remove obviously erroneous or invalid data, such as extreme outliers or data with incomplete records, and ensure that the processed data have basic consistency and availability.

[0032] Secondly, data alignment and spatial matching are performed. Since meteorological station data, ERA5 grid data and auxiliary data may come from different spatial resolutions and coordinate systems, the data are spatially aligned. Specifically, the spatial positions of meteorological station data and ERA5 grid data are matched to ensure that both are analyzed in the same spatial framework. At the same time, auxiliary data (such as surface cover type and elevation) also need to be spatially aligned with meteorological data for feature extraction and fusion.

[0033] Then, missing values are interpolated and data is smoothed. In meteorological data, an interpolation method based on data distribution characteristics is adopted. The interpolation process not only fills the missing values, but also considers the continuity and smoothness of the data to avoid introducing artificial noise. In addition, the data is smoothed to eliminate high-frequency noise and retain the main trends and characteristics of the data.

[0034] Finally, data standardization and normalization are necessary. Since the dimensions and numerical ranges of meteorological station data, ERA5 grid data, and auxiliary data may vary significantly, data standardization is required. Specifically, the values of different data sources are converted to a unified scale, for example, through linear transformation or nonlinear mapping, to make the data comparable.

[0035] In the process of meteorological data preprocessing, data screening and outlier detection, assuming that the meteorological station data is X = {x1, x2, ..., x n}, where x i represents the observation value of the i-th station.

[0036]

[0037] Where Z i represents the standardized score of the i-th site data; μ X represents the mean of the data set X; σ X Represents the standard deviation of the data set X.

[0038] Filtering rules:

[0039] If | Z i |>θ, then determine x i is an abnormal value, where θ is a preset threshold (such as θ=3).

[0040] Data alignment and spatial matching, assuming that the spatial locations of meteorological station data and ERA5 grid data are S = {(s 1x ,s 1y ),...,(s nx , s ny )} and G={(g 1x , g 1y ),...,(g mx , g my )), where s ix and s iy represents the longitude and latitude of the i-th station, g jx and g jy Indicates the longitude and latitude of the j-th grid point.

[0041]

[0042] Where, d ijrepresents the Euclidean distance between the i-th station and the j-th grid point; by minimizing d ij , spatially matching the site data with the grid data.

[0043] Missing value interpolation and data smoothing. Assuming that there are missing values in meteorological data, we use time series-based interpolation methods, such as linear interpolation or spline interpolation:

[0044]

[0045] Where x i′ represents the interpolated value; t i′ represents the i′th time point; x i′-1 and x i′+1 are the known values before and after the missing value, respectively.

[0046] Data smoothing, using the sliding average method to smooth the data:

[0047]

[0048] Where y i represents the smoothed value; k is the size of the sliding window.

[0049] For data standardization and normalization, it is assumed that the original values of meteorological station data, ERA5 grid data, and auxiliary data are X, Y, and Z, respectively, and their dimensions and value ranges are quite different.

[0050] Normalization formula:

[0051]

[0052] Where μ X 、μ Y 、μ Z are the means of X, Y, and Z respectively; σ X , σ Y , σ Z are the standard deviations of X, Y, and Z respectively.

[0053] Normalization formula:

[0054]

[0055] Where, X min 、X max are the minimum and maximum values of X respectively; Y min 、Y max are the minimum and maximum values of Y respectively; Z min , Z maxare the minimum and maximum values of Z, respectively. The above formulas cover the key steps in meteorological data preprocessing, including data screening, spatial matching, missing value interpolation, data smoothing, and standardization and normalization. These methods can significantly improve the quality and usability of meteorological data, laying a solid foundation for subsequent analysis and modeling.

[0056] In the embodiment of the present application, data screening and outlier detection can effectively identify and remove outliers in the data set through the calculation of standardized scores (Z-score); outliers may be caused by instrument failure, human recording errors or extreme weather events, and these data will interfere with subsequent analysis; after removing outliers, the data set is cleaner and more reliable, and can more realistically reflect the laws of meteorological changes; it provides a high-quality data foundation for subsequent modeling and analysis, avoiding model deviations or erroneous conclusions caused by outliers. Data alignment and spatial matching, meteorological station data, ERA5 grid data and auxiliary data (such as surface cover type and elevation) may come from different spatial resolutions and coordinate systems; by calculating Euclidean distance and performing spatial matching, ensure that all data are aligned in the same spatial framework; spatial alignment is the prerequisite for data fusion. Only by unifying data from different sources into the same spatial framework can effective feature extraction and comprehensive analysis be carried out; ensuring the spatial consistency of data in subsequent analysis, avoiding errors caused by spatial mismatch. Missing value interpolation and data smoothing: Time series interpolation methods (such as linear interpolation or spline interpolation) are used to fill missing values, and the data is smoothed using a sliding average method. Interpolation methods can effectively fill gaps in the data, while smoothing can eliminate high-frequency noise and retain the main trends of the data. Missing value interpolation ensures data integrity and avoids model failure or bias caused by missing data. Data smoothing can remove short-term random fluctuations, highlight long-term trends and main features, and make the data more suitable for modeling and prediction. Data standardization and normalization: Through standardization and normalization, the values of different data sources are converted to a unified scale. Standardization subtracts the mean and divides by the standard deviation to give the data zero mean and unit variance. Normalization scales the data to a range of 0 to 1. Because the dimensions and numerical ranges of meteorological station data, ERA5 grid data, and auxiliary data vary greatly, directly using these data may cause certain features to occupy too large a weight in the model. Standardization and normalization can eliminate the influence of dimension, making different features comparable in the model, and improving model stability and prediction accuracy.

[0057] In summary, through the above steps, meteorological data preprocessing can significantly improve the quality and availability of data, which is specifically reflected in the following aspects: by eliminating outliers, filling missing values and eliminating noise, the data is cleaner, more complete and reliable; through spatial alignment and standardization, data from different sources are analyzed under the same framework, avoiding errors caused by data inconsistency; high-quality preprocessed data can significantly improve the accuracy of subsequent modeling and analysis, enabling the model to better capture the laws of meteorological changes and provide strong support for applications such as meteorological forecasting and climate research.

[0058] S120: Use random forest regression to convert ERA5 grid data into virtual meteorological grid data; the random forest regression model learns the mapping relationship between ERA5 grid data and meteorological station data, and generates virtual meteorological grid data with similar accuracy to meteorological station data. Figure 3 ;

[0059] Optionally, in some embodiments, the process of generating virtual weather grid data with an accuracy close to that of weather station data comprises the following steps:

[0060] First, the meteorological elements (such as temperature, humidity, and air pressure) in the ERA5 grid data are decomposed at multiple levels to extract spatiotemporal features at different scales. This is done in a step-by-step manner, from coarse granularity to fine granularity, to gradually capture the complex relationships between the ERA5 grid data and the site data. The decomposition not only focuses on the local changes of a single meteorological element, but also explores the implicit spatiotemporal patterns through the interaction between the features.

[0061] Secondly, based on the feature decomposition, the association relationship between ERA5 grid data and site data is dynamically constructed through a nonlinear mapping mechanism based on a tree structure. Each grid point in the ERA5 grid data is matched with the site data in a layer-by-layer recursive manner, and the influence weight of the grid point on the site data is dynamically adjusted according to the distribution characteristics of the site data. The dynamic weight allocation mechanism can adaptively capture the nonlinear changes in the data and ensure the accuracy of the mapping relationship.

[0062] Finally, after the mapping relationship is constructed, the feature space of the ERA5 grid data is optimized; the key features in the ERA5 grid data are integrated with the site data through feature fusion to generate a high-precision virtual meteorological grid data; this virtual data not only retains the spatiotemporal continuity of the ERA5 grid data, but also significantly improves the consistency with the site data through feature optimization.

[0063] It should be noted that, in the embodiment, the process of extracting spatiotemporal features of different scales specifically includes the following steps:

[0064] The ERA5 grid data is divided into multiple local areas, and the basic statistical characteristics of meteorological elements (such as mean, variance, extreme value, etc.) are extracted in each area.

[0065] The features of multiple local areas are integrated to form coarse-grained spatiotemporal features; the transition areas and change trends of meteorological elements in space are identified through gradient analysis, and medium-grained spatiotemporal features are extracted; through local extreme value detection, small fluctuations and abnormal changes of meteorological elements in space are identified, and fine-grained spatiotemporal features are extracted.

[0066] Identify the correlation between different meteorological elements; for example, analyze the interactions between elements such as temperature and humidity, air pressure and wind speed, and extract the correlation patterns between features; based on the feature correlation analysis, construct the spatiotemporal patterns of meteorological elements through spatiotemporal sequence analysis; for example, analyze the temporal evolution laws and spatial distribution characteristics of meteorological elements, and extract representative spatiotemporal patterns.

[0067] The spatiotemporal features of different scales are fused to form a set of feature sets; the feature sets not only retain the macro-changing trends of meteorological elements, but also include local detailed features and implicit spatiotemporal patterns.

[0068] It should be noted that, in the embodiment, the process of dynamically building the association relationship between ERA5 grid data and site data specifically includes the following steps:

[0069] Based on the feature decomposition, the association relationship between grid data and site data is dynamically constructed through layer-by-layer recursion of the tree structure. Specifically, each layer of the tree corresponds to a specific spatiotemporal scale, progressing from coarse-grained to fine-grained. In each layer, each grid point of the grid data is matched with the site data, and the influence weight of the grid point on the site data is dynamically adjusted according to the distribution characteristics of the site data.

[0070] In each layer of the tree structure, the distribution of weights is not fixed, but is dynamically adjusted according to the distribution characteristics of the site data and the spatiotemporal characteristics of the grid data. For example, when the site data in a certain area is relatively dense, the weight of the influence of the grid on the site data will increase accordingly; while in areas with sparse site data, the weight will decrease.

[0071] Among them, when dynamically building the association between ERA5 grid data and site data, the allocation of weights is a key step. In order to ensure that the dynamic adjustment of weights can accurately reflect the distribution characteristics of site data and the spatiotemporal characteristics of grid data, the dynamic adjustment of weights is achieved through the following formula:

[0072]

[0073] Where Wi″，j″，k′ Indicates the weight of the grid point (i″, j″) to the site data in the k′th layer; S m represents the observation value of the mth station; represents the spatial distance between the grid point (i″, j″) and the mth site; represents the time difference between the grid point (i″, j″) and the mth station; represents the spatial scale parameter of the k′th layer, which controls the influence of spatial distance on the weight; represents the time scale parameter of the k′th layer, which controls the impact of time difference on the weight; N represents the total number of sites.

[0074] Spatial distance weight formula:

[0075]

[0076] Where w d，i″，j″，m represents the spatial distance weight between the grid point (i″, j″) and the mth site;

[0077] Time difference weight formula:

[0078]

[0079] Where w t,i″，j″，m Represents the time difference weight between grid point (i″, j″) and the mth site.

[0080] Combining the weights of spatial distance and time difference, we get the final weight formula:

[0081]

[0082] Where W i″,j″,k′ represents the weight of the grid point (i″, j″) to the site data in the k′th layer; w d，i″，j″，m represents the spatial distance weight between the grid point (i″, j″) and the mth site; w t,i″，j″，m Represents the time difference weight between grid point (i″, j″) and the mth site.

[0083] In order to further dynamically adjust σ k′ and τ k′ , expressed using the following formula:

[0084]

[0085] Where, σ k′ represents the spatial scale parameter of the k′th layer; τ k′The k′th layer's time scale parameter is represented by σ0, the initial spatial scale parameter, and τ0, the initial time scale parameter. α and β are constants that control the decay rate of these parameters. Using these formulas, we can dynamically build the relationship between ERA5 grid data and site data, and dynamically adjust the weight distribution based on the distribution characteristics of the site data and the spatiotemporal characteristics of the grid data.

[0086] It should be noted that, in the embodiment, the process of optimizing the feature space of ERA5 grid data specifically includes the following steps:

[0087] The feature space of ERA5 grid data is reconstructed at multiple levels. The reconstruction of each level is based on the feedback of the previous level, and the resolution of the feature space is gradually refined. At the global level, attention is paid to the overall distribution pattern of meteorological elements; at the local level, the focus is on the deviation area between the station data and the grid data.

[0088] The key features of the ERA5 grid data are matched with the site data, and the weight of each grid point in the grid data is optimized according to the distribution characteristics of the site data.

[0089] After the dynamic feature alignment is completed, the feature distribution of the grid data is dynamically adjusted to achieve a smooth transition in space while highly matching the local features of the site data. Through multiple iterative optimizations, a high-precision virtual meteorological grid dataset is ultimately generated.

[0090] Through multi-level reconstruction, dynamic feature alignment, feature fusion, and iterative optimization, a novel feature space optimization mechanism has been constructed. This mechanism not only differs from existing technical methods but also possesses significant inventiveness, meeting the definition of inventiveness under patent law. The virtual meteorological grid data generated through this process significantly improves consistency with station data while preserving the spatiotemporal continuity of the ERA5 grid data.

[0091] In an embodiment of the present application, multi-level decomposition and spatiotemporal feature extraction, by dividing the ERA5 grid data into multiple local areas and extracting spatiotemporal features of different scales (such as mean, variance, extreme values, etc.), can fully capture the macro trends and micro changes of meteorological elements; gradient analysis and local extreme value detection further refine the accuracy of feature extraction, ensuring the integrity of data at different spatiotemporal scales. It provides high-quality feature input for mapping relationship construction, significantly improves the model's learning ability for complex meteorological patterns, and lays the foundation for generating high-precision virtual meteorological grid data. The association relationship between grid data and site data is dynamically constructed, and the association relationship between ERA5 grid data and site data is dynamically constructed through layer-by-layer recursion of the tree structure. The dynamic adjustment mechanism of weights (based on spatial distance and time difference) can adaptively reflect the distribution characteristics of site data and ensure the accuracy of the mapping relationship. It solves the problem of insufficient precision caused by fixed weight allocation in traditional methods, and significantly improves the consistency between virtual meteorological grid data and site data, especially in areas where site data are unevenly distributed. Multi-level reconstruction and optimization of the feature space: This involves multi-level reconstruction of the feature space of ERA5 grid data, gradually refining the feature resolution. Dynamic feature alignment and weight optimization are used to eliminate the deviation between station data and grid data. The feature fusion mechanism further integrates the key features of the ERA5 grid data with the station data to generate high-precision virtual meteorological grid data. This ensures the spatial continuity and temporal stability of the virtual meteorological grid data, while significantly improving its consistency with the station data, providing more reliable data support for meteorological research and applications. Dynamic weight allocation and parameter optimization: By dynamically adjusting spatial and temporal scale parameters, the dynamic weight allocation ensures that the weight allocation accurately reflects the distribution characteristics of the station data and the spatiotemporal characteristics of the grid data. It can adaptively capture nonlinear changes in the data, avoiding overfitting or underfitting. The dynamic weight allocation mechanism significantly improves the model's generalization and prediction accuracy, particularly under complex meteorological conditions, enabling better simulation of the spatiotemporal variations of meteorological elements. Feature fusion and iterative optimization* dynamically adjusts the feature distribution of grid data through multiple iterations, ensuring a smooth spatial transition and a close match with the local characteristics of the station data. An adaptive feedback mechanism continuously optimizes feature fusion parameters to ensure optimal consistency between virtual meteorological grid data and station data. This feature fusion and iterative optimization mechanism significantly improves the accuracy and reliability of virtual meteorological grid data, providing high-quality data support for applications such as weather forecasting and climate research.

[0092] In summary, this embodiment constructs a new feature space optimization mechanism through multi-level decomposition, dynamic weight allocation, feature fusion and iterative optimization. It can significantly improve the consistency between virtual meteorological grid data and site data, while retaining the spatiotemporal continuity of ERA5 grid data. It not only solves the problem of insufficient accuracy in traditional methods, but also has significant creativity and practicality; by generating high-precision virtual meteorological grid data, it can provide more reliable data support for meteorological forecasting, disaster warning, climate research and other fields, and has important scientific value and application prospects. Each step jointly realizes the generation of high-precision virtual meteorological grid data through multi-level decomposition, dynamic weight allocation, feature fusion and iterative optimization. This process not only significantly improves the consistency of the data, but also retains the spatiotemporal continuity of ERA5 grid data, providing high-quality data support for meteorological research and applications.

[0093] S130: Train a Dynamic Spatiotemporal Decay Network (DSTDNet). The Dynamic Spatiotemporal Decay Network learns spatiotemporal features through 3D convolution, spatiotemporal attention mechanism, and GRU module. By inputting virtual meteorological grid data and real-time station data, the Dynamic Spatiotemporal Decay Network learns the spatiotemporal decay rate of severe convective weather and establishes the relationship between spatiotemporal features and forecast results. The structure of the Dynamic Spatiotemporal Decay Network (DSTDNet) is shown in the attached figure. Figure 4 .

[0094] Optionally, in some embodiments, the process of establishing the relationship between the spatiotemporal features and the forecast results specifically includes the following steps:

[0095] The virtual meteorological grid data and real-time station data are preliminarily processed through three-dimensional convolution operations to extract the spatiotemporal characteristics of meteorological elements; three-dimensional convolution can simultaneously capture the changing patterns of meteorological data in spatial and temporal dimensions and form a preliminary feature representation.

[0096] Based on the preliminary feature extraction, the spatiotemporal attention mechanism is introduced to dynamically assign feature weights to different spatiotemporal regions; by calculating the spatial correlation and temporal continuity of meteorological elements, the regions and moments that have a key impact on severe convective weather forecasts are identified.

[0097] The attenuation law of meteorological elements in the spatiotemporal dimensions is learned through the gated recurrent unit (GRU) module; the GRU module captures the attenuation characteristics of meteorological elements at different time scales through the dynamic update mechanism of time series and generates a dynamic attenuation rate.

[0098] Based on the dynamic attenuation rate, the spatiotemporal features and forecast results are associated and modeled through the feature fusion mechanism; the extracted spatiotemporal features are weightedly fused with the dynamic attenuation rate to generate the final forecast results.

[0099] It should be noted that, in the embodiment, in the process of establishing the relationship between the spatiotemporal characteristics and the forecast results, the calculation formulas involved in each calculation step are as follows:

[0100] 3D convolution operation extracts spatiotemporal features:

[0101]

[0102] Where, F s′，t′ represents the spatiotemporal features extracted at the spatial position (s′, t′) and time dimension k″; W i″′,j″′,k″ represents the weight of the 3D convolution kernel at position (i′′′, j″′, k″); X s′+i″′-1，t′+j″′-1，k″ represents the value of the input meteorological data at the spatial position (s′+i″′-1, t′+j″′-1) and the time dimension k″; b represents the bias term; (H, W, T\) represents the size of the convolution kernel in the height, width, and time dimensions. Through the three-dimensional convolution operation, the changing patterns of meteorological data in the spatial and temporal dimensions are captured simultaneously to form a preliminary spatiotemporal feature representation.

[0103] The spatiotemporal attention mechanism dynamically assigns feature weights:

[0104]

[0105] Where, α s′，t′，k″ represents the attention weight at the spatial position (s′, t′) and the time dimension k″; represents the query vector, which represents the characteristics of the current spatiotemporal position; K s′，t′，k″ represents the key vector, which represents the characteristics of other spatiotemporal locations; d represents the dimension of the characteristic vector; by calculating the spatial correlation and temporal continuity of meteorological elements, the characteristic weights of different spatiotemporal regions are dynamically assigned, and the regions and moments that have a key impact on severe convective weather forecasts are identified.

[0106] The GRU module learns the attenuation law of meteorological elements:

[0107] z t′ =σ(W z ·[h t′-1 , x t′ ]+b z )

[0108] r t′ =σ(W r ·[h t′-1 , x t′]+b r )

[0109]

[0110] Where z t′ represents the update gate, which controls the retention ratio of historical information; r t′ Represents the reset gate, which controls the forgetting ratio of historical information; represents the candidate hidden state, which represents the potential state at the current moment; h t′ Indicates the hidden state at the current moment; W z , W r , W h represents the weight matrix; b z , b r , b h denotes the bias term; σ denotes the Sigmoid activation function; and ⊙ denotes element-by-element multiplication. Through the GRU module's time series dynamic update mechanism, the attenuation characteristics of meteorological elements at different time scales are captured and a dynamic attenuation rate is generated.

[0111] The feature fusion mechanism generates the final forecast results:

[0112]

[0113] Where, Y represents the final forecast result; α s′，t′，k″ represents the spatiotemporal attention weight; F s′，t′，k″ represents the extracted spatiotemporal features; D s′，t′，k″ represents the dynamic decay rate; b y Represents the bias term. Through the feature fusion mechanism, the extracted spatiotemporal features are weightedly fused with the dynamic attenuation rate to generate the final forecast result.

[0114] The above formula builds a complete model of the relationship between spatiotemporal features and forecast results through three-dimensional convolution operations, spatiotemporal attention mechanisms, GRU modules, and feature fusion mechanisms. Each formula incorporates complex calculations and rich character meanings, ensuring the model's accuracy and flexibility.

[0115] S140: Apply the learned spatiotemporal attenuation rate to the severe convection prediction data, use the spatiotemporal attenuation rate to correct the severe convective weather forecast results output by the severe convection numerical prediction model, revise the results, and output accurate severe convective weather forecast results.

[0116] Optionally, in some embodiments, the process of correcting the severe convective weather forecast result output by the severe convective numerical prediction model using the spatiotemporal attenuation rate specifically includes the following steps:

[0117] The spatiotemporal attenuation rate reflects the attenuation characteristics of severe convective weather at different time and space scales; through the spatiotemporal attenuation rate, the "time lag" or "time advance" phenomenon that may exist in the forecast results can be identified; the spatial distribution of severe convective weather often shows non-uniformity, and the spatiotemporal attenuation rate can capture the characteristics of this spatial variation; by combining the attenuation rate with the spatial distribution of the forecast results, the spatial deviation of the model can be corrected; after the initial adjustment of the spatiotemporal attenuation rate, the forecast results have been optimized in the time and space dimensions.

[0118] The forecast results after preliminary adjustments need to be further optimized through a multi-level feedback mechanism. By comparing the forecast results after preliminary adjustments with real-time station observation data, errors in local areas can be identified; local errors can be corrected using the spatiotemporal attenuation rate. For example, if the precipitation in a certain area is overestimated, the spatiotemporal attenuation rate will dynamically adjust the forecast value according to the attenuation characteristics of the area to make it closer to actual observations.

[0119] Severe convective weather has regional synergy, that is, weather changes in a certain area may affect neighboring areas. By analyzing the synergistic relationship between regions, the forecast results can be further optimized; if the severe convective weather in a certain area decays faster, while the decay rate in the neighboring area is slower, the spatiotemporal attenuation rate will dynamically adjust the forecast results according to the synergistic relationship between regions.

[0120] Based on local and regional optimization, the global forecast results are evaluated to identify possible systematic deviations; if the overall forecast results show a certain trend deviation (such as generally high precipitation), the spatiotemporal attenuation rate is dynamically balanced and adjusted according to the global attenuation characteristics; during the forecast process, real-time site data will be continuously updated; each update of real-time data will trigger a new feedback loop; after the above-mentioned multi-level feedback mechanism optimization, the forecast results have been significantly improved in time, space and global scope; the final output can not only accurately reflect the spatiotemporal evolution of severe convective weather, but also dynamically adapt to real-time meteorological changes, providing more reliable decision-making support for disaster prevention and mitigation.

[0121] It should be noted that, in the embodiment, through the organic combination of spatiotemporal attenuation rate and multi-level feedback mechanism, dynamic optimization of forecast results is achieved, breaking through the limitations of traditional methods and significantly improving the accuracy and practicality of the forecast.

[0122] In the embodiments of the present application, the above is an introduction to the method embodiment. The following further illustrates the solution of the present disclosure through an apparatus embodiment.

[0123] Figure 5 FIG. 2 shows a block diagram of a severe convective weather forecast result correction system 200 according to an embodiment of the present disclosure. Figure 5As shown, the apparatus 200 includes:

[0124] The data acquisition module 210 is used to collect meteorological station data, ERA5 reanalysis meteorological data, and auxiliary data such as surface cover type and elevation; preprocess the meteorological station data, meteorological data and auxiliary data; preprocess missing value interpolation and standardization, etc.

[0125] The random forest regression module 220 is used to convert the ERA5 grid data into virtual meteorological grid data using random forest regression. The random forest regression model learns the mapping relationship between the ERA5 grid data and the meteorological station data to generate virtual meteorological grid data with a precision close to that of the meteorological station data.

[0126] The dynamic spatiotemporal decay network module 230 is used to train a dynamic spatiotemporal decay network (DSTDNet). The dynamic spatiotemporal decay network learns spatiotemporal features through 3D convolution, spatiotemporal attention mechanism and GRU module. By inputting virtual meteorological grid data and real-time station data, the dynamic spatiotemporal decay network learns the spatiotemporal decay rate of severe convective weather and establishes the relationship between spatiotemporal features and forecast results.

[0127] The result output module 240 is used to apply the learned spatiotemporal attenuation rate to the severe convection prediction data, use the spatiotemporal attenuation rate to correct the severe convective weather forecast results output by the severe convection numerical prediction model, revise the results, and output accurate severe convective weather forecast results.

[0128] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0129] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

[0130] Figure 6 A schematic block diagram of an electronic device 300 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0131] The electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a ROM 302 or a computer program loaded from a storage unit 308 into a RAM 303. The RAM 303 may also store various programs and data required for the operation of the electronic device 300. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An I / O interface 305 is also connected to the bus 304.

[0132] Multiple components in the electronic device 300 are connected to the I / O interface 305, including an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a magnetic disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the electronic device 300 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0133] The computing unit 301 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as the method for risk zoning for aircraft flights. For example, in some embodiments, the method for risk zoning for aircraft flights can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the method for risk zoning for aircraft flights described above can be performed. Alternatively, in other embodiments, the computing unit 301 may be configured in any other appropriate manner (for example, by means of firmware) to execute the method for risk zoning for aircraft flight.

[0134] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0135] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0136] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0137] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0138] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0139] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0140] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0141] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A method for correcting severe convective weather forecast results, characterized in that: include: Collect meteorological station data, ERA5 reanalysis meteorological data, and auxiliary data on land cover type and elevation; Preprocessing of meteorological station data, meteorological data and auxiliary data; Preprocessing missing value imputation and standardization; The random forest regression model learns the mapping relationship between ERA5 grid data and meteorological station data, generating virtual meteorological grid data with similar accuracy to meteorological station data; A dynamic spatiotemporal decay network is trained. The network learns spatiotemporal features through 3D convolution, spatiotemporal attention mechanism, and GRU module. By inputting virtual meteorological grid data and real-time station data, the network learns the spatiotemporal decay rate of severe convective weather and establishes the relationship between spatiotemporal features and forecast results. The learned spatiotemporal attenuation rate is applied to the severe convection prediction data, and the spatiotemporal attenuation rate is used to correct the severe convective weather forecast results output by the severe convection numerical prediction model, and the results are corrected to output accurate severe convective weather forecast results.

2. The method according to claim 1, characterized in that The meteorological station data includes wind speed and precipitation, and the meteorological data changes in temperature, air pressure, humidity, wind and cloud cover are analyzed.

3. The method according to claim 1, characterized in that The process of preprocessing the meteorological station data, meteorological data and auxiliary data includes the following steps: First, raw data were collected from meteorological stations, ERA5 reanalysis data, and land cover type and elevation auxiliary data sources; Secondly, data alignment and spatial matching are performed. Since meteorological station data, ERA5 grid data, and auxiliary data may come from different spatial resolutions and coordinate systems, the data are spatially aligned. Specifically, the spatial positions of meteorological station data and ERA5 grid data are matched. At the same time, the auxiliary data are also spatially aligned with the meteorological data. Then, missing value interpolation and data smoothing are performed. In meteorological data, an interpolation method based on data distribution characteristics is used. The interpolation process fills in missing values, performs smoothing, and eliminates high-frequency noise. Finally, data standardization and normalization convert the values of different data sources into a unified scale range.

4. The method according to claim 1, wherein The process of generating virtual meteorological grid data with similar accuracy to meteorological station data includes the following steps: First, the meteorological elements in the ERA5 grid data are decomposed into multiple levels to extract spatiotemporal features at different scales. This is done by progressively decomposing the meteorological elements in the ERA5 grid data from coarse granularity to fine granularity, gradually capturing the complex relationships between the ERA5 grid data and the station data. Secondly, based on the eigendecomposition, a tree-based nonlinear mapping mechanism is used to dynamically build the association between ERA5 grid data and station data. Each grid point in the ERA5 grid data is matched with the station data in a layer-by-layer recursive manner, and the influence weight of the grid point on the station data is dynamically adjusted according to the distribution characteristics of the station data. Finally, after the mapping relationship is constructed, the feature space of ERA5 grid data is optimized; Through feature fusion, the key features in the ERA5 grid data are integrated with the site data to generate a high-precision virtual meteorological grid data.

5. The method according to claim 4, characterized in that The process of extracting spatiotemporal features of different scales includes the following steps: The ERA5 grid data is divided into multiple local areas, and the basic statistical characteristics of meteorological elements are extracted in each area; The features of multiple local areas are integrated to form coarse-grained spatiotemporal features. The transition areas and change trends of meteorological elements in space are identified through gradient analysis to extract medium-grained spatiotemporal features. The small fluctuations and abnormal changes of meteorological elements in space are identified through local extreme value detection to extract fine-grained spatiotemporal features. Identify the correlation between different meteorological elements; based on the feature correlation analysis, construct the spatiotemporal pattern of meteorological elements through spatiotemporal sequence analysis; The spatiotemporal features of different scales are fused to form a set of feature sets.

6. The method according to claim 5, characterized in that The process of dynamically building the association relationship between ERA5 grid data and site data includes the following steps: Based on feature decomposition, the relationship between grid data and station data is dynamically constructed through layer-by-layer recursion of a tree structure. Each layer of the tree corresponds to a specific spatiotemporal scale, progressing from coarse granularity to fine granularity. In each layer, each grid point of the grid data is matched with the station data, and the influence weight of the grid point on the station data is dynamically adjusted based on the distribution characteristics of the station data. In each layer of the tree structure, the distribution of weights is not fixed, but is dynamically adjusted according to the distribution characteristics of the site data and the spatiotemporal characteristics of the grid data.

7. The method according to claim 6, characterized in that When dynamically building the association between ERA5 grid data and site data, the dynamic adjustment of weights is achieved through the following formula: Where W i″，j″，k′ Indicates the weight of the grid point (i″, j″) to the site data in the k′th layer; S m represents the observation value of the mth station; represents the spatial distance between the grid point (i″, j″) and the mth site; represents the time difference between the grid point (i″, j″) and the mth station; represents the spatial scale parameter of the k′th layer, which controls the influence of spatial distance on the weight; represents the time scale parameter of the k′th layer, which controls the influence of time difference on weight; N represents the total number of sites; Spatial distance weight formula: Where w d，i″，j″，m represents the spatial distance weight between the grid point (i″, j″) and the mth site; Time difference weight formula: Where w t，i″，j″，m represents the time difference weight between the grid point (i″, j″) and the m-th site; Combining the weights of spatial distance and time difference, we get the final weight formula: Where W i″，j″，k′ represents the weight of the grid point (i″, j″) to the site data in the k′th layer; w d，i″，j″，m represents the spatial distance weight between the grid point (i″, j″) and the mth site; w t，i″，j″，m represents the time difference weight between the grid point (i″, j″) and the m-th site; To further dynamically adjust σ k′ and τ k′ , expressed using the following formula: Where, σ k′ represents the spatial scale parameter of the k′th layer; τ k′ represents the time scale parameter of the k′th layer; σ0 represents the initial spatial scale parameter; τ0 represents the initial time scale parameter; α and β represent constants that control the decay rate of the parameters.

8. The method according to claim 6, characterized in that The process of optimizing the feature space of ERA5 grid data includes the following steps: The feature space of ERA5 grid data is reconstructed at multiple levels. Each level of reconstruction is based on the feedback from the previous level, gradually refining the resolution of the feature space. At the global level, attention is paid to the overall distribution of meteorological elements; at the local level, the focus is on the deviation areas between the station data and the grid data. Match the key features of ERA5 grid data with the site data, and optimize the weight of each grid point in the grid data according to the distribution characteristics of the site data; After the dynamic feature alignment is completed, the feature distribution of the grid data is dynamically adjusted to achieve a smooth transition in space while highly matching the local features of the site data. Through multiple iterative optimizations, a high-precision virtual meteorological grid dataset is ultimately generated.

9. The method according to claim 1, characterized in that The process of establishing the relationship between the spatiotemporal characteristics and the forecast results includes the following steps: The virtual meteorological grid data and real-time station data are preliminarily processed through three-dimensional convolution operations to extract the spatiotemporal characteristics of meteorological elements. Three-dimensional convolution can simultaneously capture the changing patterns of meteorological data in spatial and temporal dimensions, forming a preliminary feature representation. Based on the preliminary feature extraction, a spatiotemporal attention mechanism is introduced to dynamically assign feature weights to different spatiotemporal regions. By calculating the spatial correlation and temporal continuity of meteorological elements, the regions and moments that have a key impact on severe convective weather forecasts are identified. The Gated Recurrent Unit (GRU) module is used to learn the attenuation patterns of meteorological elements in the spatiotemporal dimension. The GRU module captures the attenuation characteristics of meteorological elements at different time scales and generates dynamic attenuation rates through the dynamic update mechanism of time series. Based on the dynamic attenuation rate, the spatiotemporal features and forecast results are associated and modeled through the feature fusion mechanism; the extracted spatiotemporal features are weightedly fused with the dynamic attenuation rate to generate the final forecast results.

10. A severe convective weather forecast result correction system, characterized in that: include: Data acquisition module, used to collect meteorological station data, ERA5 reanalysis meteorological data, and surface cover type and elevation auxiliary data; Preprocessing of meteorological station data, meteorological data and auxiliary data; Random forest regression module, used to convert ERA5 grid data into virtual meteorological grid data using random forest regression; The dynamic spatiotemporal decay network module is used to train a dynamic spatiotemporal decay network. The dynamic spatiotemporal decay network learns spatiotemporal features through 3D convolution, spatiotemporal attention mechanism, and GRU module. By inputting virtual meteorological grid data and real-time station data, the dynamic spatiotemporal decay network learns the spatiotemporal decay rate of severe convective weather and establishes the relationship between spatiotemporal features and forecast results. The result output module is used to apply the learned spatiotemporal attenuation rate to the severe convection prediction data, use the spatiotemporal attenuation rate to correct the severe convective weather forecast results output by the severe convection numerical prediction model, revise the results, and output accurate severe convective weather forecast results.

Images