A delay-aware hierarchical spatio-temporal weather forecasting method and system

Abstract

The application discloses a delay-aware hierarchical spatio-temporal weather prediction method and system, and relates to the technical field of weather prediction. The method comprises the following steps: dividing sites into multiple regions according to spatial positions; for any two sites in the same region, calculating cross-correlation scores under candidate delays according to observation data, constructing a propagation delay prior based on physical laws, obtaining data delay probability distribution and physical delay probability distribution respectively, and obtaining local propagation delay and its weight between site pairs after fusion; constructing a regional delay, and correcting and aggregating the site representation in the region in time to obtain a virtual site representation of the next layer, thereby obtaining a hierarchical weather representation; and on this basis, modeling spatial propagation relationships and time evolution laws of different levels, fusing spatio-temporal characteristics of each layer, and obtaining a prediction result of meteorological observation data, so that the real state of weather in different spatial ranges can be more accurately described.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting technology, and in particular to a delayed-sensing hierarchical spatiotemporal weather forecasting method and system. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Spatiotemporal weather forecasting is an important research direction at the intersection of meteorological information processing and artificial intelligence. Its goal is to use historical meteorological observation data to predict meteorological elements such as temperature, humidity, wind speed, and air pressure within a certain future time range. Accurate weather forecasts are of great significance for disaster prevention and mitigation, agricultural production, transportation, energy dispatching, ecological monitoring, and urban management.

[0004] With the continuous accumulation of meteorological station, remote sensing observation, and reanalysis data, deep learning-based weather forecasting methods have gradually become a research hotspot. Existing methods typically use models such as recurrent neural networks, convolutional neural networks, graph neural networks, or Transformers to jointly model the temporal and spatial dependencies in meteorological data, thereby improving forecast accuracy in complex meteorological scenarios.

[0005] However, real weather systems do not change synchronously in all spatial locations, but rather spread gradually between different geographical regions and stations along with atmospheric motion. The arrival time of the same meteorological process at different stations often varies, and this time difference is affected by factors such as station distance, topographic conditions, underlying surface features, and the direction of weather system propagation.

[0006] Existing weather algorithms have significant shortcomings, specifically in the following aspects: (1) Failure to fully consider the impact of time delay on spatial correlation leads to misjudgment of correlation: Existing technologies generally ignore the profound impact of propagation time delay on the spatial correlation characteristics between meteorological nodes. This makes it easy for models to misjudge the delayed correlations naturally formed by atmospheric propagation physical processes as weak correlations or even no correlations when extracting spatiotemporal features. This misjudgment of the real physical connections seriously weakens the model's ability to express the features of complex spatiotemporal dynamic patterns.

[0007] (2) Lack of temporal misalignment correction mechanism leads to distortion of multi-scale feature representation: Existing methods fail to effectively align and correct temporal misalignment caused by propagation between different stations when constructing and characterizing the multi-scale (such as local, regional, and global) hierarchical structure of weather systems. This results in the direct and rigid fusion of observational data at different evolution and propagation phases. This misaligned fusion causes the generated high-level regional feature representation to be mixed with meteorological states at different time phases, ultimately leading to inaccurate representation of macroscopic spatiotemporal features at the regional and even global levels, making it difficult to reflect the true evolution of weather systems. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes a hierarchical spatiotemporal weather prediction method and system based on delayed perception. It introduces propagation delay modeling to correct the time misalignment caused by weather propagation at different stations, thereby more accurately depicting the true state of weather in different spatial ranges.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a delayed-sensing hierarchical spatiotemporal weather prediction method, comprising: Based on the spatial location of each meteorological station, meteorological observation data, and the corresponding observation time, a weather representation of the station is constructed, and all stations are divided into multiple regions according to their spatial location. For any two stations in the same area, the cross-correlation score under the candidate delay is calculated based on meteorological observation data, and a propagation delay prior based on physical laws is constructed. The data delay probability distribution and physical delay probability distribution are obtained respectively and then fused to obtain the local propagation delay and its weight between the station pairs. The regional delay is obtained based on the local propagation delay and its weight between the stations. The time misalignment of the weather representations of the stations in the region is corrected based on the regional delay. The corrected weather representations of the stations in the same region are aggregated and used as the virtual weather representations of the next layer until there is only one virtual weather representation in the current layer. This process is repeated to obtain a hierarchical weather representation. For each layer in the hierarchical weather representation, spatial features are extracted after sorting the stations in the region according to regional delay. Temporal features are extracted after frequency band decomposition and reconstruction of the meteorological observation data of each station. The spatial and temporal features of each layer are then fused to obtain the prediction results of the meteorological observation data.

[0010] As an alternative implementation method, the process of constructing a propagation delay prior based on physical laws includes: Calculate the ideal propagation delay between site pairs: ; With ideal propagation delay Construct candidate delays centered on [the target group]. Gaussian physical prior distribution: ; in, For the site With the site The spherical distance between them The sampling interval is... The propagation speed of meteorological observation data along the direction of the line connecting the stations; For the propagation delay prior, let the propagation delay be represented. The possibility; Let Variance be the variance.

[0011] As an alternative implementation method, the data latency probability distribution and the physical latency probability distribution are respectively: Candidate delay Limited to a preset range Inside, only retain The corresponding cross-correlation score and , , This represents the total time of meteorological observation data collection; Expand the candidate delay to a signed interval. Extend the cross-correlation score and propagation delay prior to the signed axis: ; in, Indicates site Weather phenomena relative to In advance, that is, the direction of transmission is ; Indicates site Weather phenomena relative to Lag, meaning the direction of propagation is ; After normalizing the cross-correlation score and the propagation delay prior, the data delay probability distribution is obtained. and physical delay probability distribution .

[0012] As an alternative implementation, the process of obtaining the local propagation delay and its weight between site pairs includes: Based on data latency probability distribution Calculate the confidence level of normalized entropy ; based on Calculate the fusion gating coefficient: ;in, This represents the upper limit of the maximum proportion of the physical delay probability distribution in the fusion. Both are hyperparameters used to adjust the sensitivity of the fusion gating coefficients to changes in confidence level. Based on the fusion gating coefficient The data delay probability distribution and the physical delay probability distribution are fused to obtain a fused distribution. The maximum value of the fused distribution is taken, and the delay that produces the maximum value is taken as the local propagation delay, with the maximum value as the weight.

[0013] As an alternative implementation, based on the local propagation delay between site pairs and their weights The process of obtaining regional latency includes: for each station within the region Introducing regional delay And construct propagation consistency constraints to make Approaching Solve the following weighted least squares problem to obtain the region-level delay. : ; Among them, constraints Used to eliminate the non-uniqueness of global translation.

[0014] As an alternative implementation method, the process of extracting time features includes: For the first Sites in the layer The observed data were subjected to wavelet decomposition, and the different frequency band components were modeled separately and the time series was reconstructed using inverse wavelet transform; among them, wavelet decomposition The number of times is: ; in, and These represent the minimum and maximum allowed number of decompositions, respectively; This represents the total number of floors.

[0015] Secondly, the present invention provides a delayed-sensing hierarchical spatiotemporal weather prediction system, comprising: The segmentation module is configured to construct a weather representation for each meteorological station based on its spatial location, meteorological observation data, and corresponding observation time, and to divide all stations into multiple regions according to their spatial location. The delay estimation module is configured to calculate the cross-correlation score under candidate delay based on meteorological observation data for any two stations in the same area, and to construct a propagation delay prior based on physical laws. In this way, the data delay probability distribution and the physical delay probability distribution are obtained respectively, and after fusion, the local propagation delay between the station pairs and their weights are obtained. The hierarchical modeling module is configured to obtain the regional delay based on the local propagation delay and its weight between the station pairs, perform time misalignment correction on the weather representation of the stations in the region based on the regional delay, aggregate the corrected weather representation of the stations in the same region, and use it as the virtual weather representation of the next layer, until there is only one virtual weather representation in the current layer, thereby obtaining the hierarchical weather representation. The spatiotemporal modeling module is configured to extract spatial features from stations within the region after sorting them by regional delay for each layer in the hierarchical weather representation, extract temporal features after frequency band decomposition and reconstruction of meteorological observation data from each station, and fuse the spatial and temporal features of each layer to obtain the prediction results of meteorological observation data.

[0016] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0017] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0018] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a hierarchical spatiotemporal weather forecasting method and system based on delay perception. By explicitly introducing propagation delay modeling during the weather forecasting process, it effectively corrects temporal misalignments caused by weather propagation at different stations, thereby effectively modeling real-world delay correlations. Simultaneously, it avoids the accumulation of propagation misalignments layer by layer during hierarchical aggregation, thus more accurately depicting the true state of weather systems in different spatial ranges. Furthermore, by collaboratively estimating propagation delay using data and physical laws, the delay estimate reflects both the actual changes in the observed data and is constrained by physical laws such as propagation distance and speed, thereby improving the stability of the delay estimate. Ultimately, this enhances the accuracy, robustness, and interpretability of spatiotemporal weather forecasting in complex meteorological scenarios.

[0020] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 Flowchart of the delayed-sensing hierarchical spatiotemporal weather prediction method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of local propagation delay estimation for data-physical cooperation provided in Embodiment 1 of the present invention; Figure 3 This is a flowchart of the hierarchical representation construction for region delay awareness provided in Embodiment 1 of the present invention; Figure 4 This is a flowchart of hierarchical spatiotemporal modeling provided in Embodiment 1 of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “including”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0027] Example 1 This embodiment proposes a delayed-sensing hierarchical spatiotemporal weather prediction method, including the following steps: Based on the spatial location of each meteorological station, meteorological observation data, and the corresponding observation time, a weather representation of the station is constructed, and all stations are divided into multiple regions according to their spatial location. For any two stations in the same area, the cross-correlation score under the candidate delay is calculated based on meteorological observation data, and a propagation delay prior based on physical laws is constructed. The data delay probability distribution and physical delay probability distribution are obtained respectively and then fused to obtain the local propagation delay and its weight between the station pairs. The regional delay is obtained based on the local propagation delay and its weight between the stations. The time misalignment of the weather representations of the stations in the region is corrected based on the regional delay. The corrected weather representations of the stations in the same region are aggregated and used as the virtual weather representations of the next layer until there is only one virtual weather representation in the current layer. This process is repeated to obtain a hierarchical weather representation. For each layer in the hierarchical weather representation, spatial features are extracted after sorting the stations in the region according to regional delay. Temporal features are extracted after frequency band decomposition and reconstruction of the meteorological observation data of each station. The spatial and temporal features of each layer are then fused to obtain the prediction results of the meteorological observation data.

[0028] The following is combined with Figures 1-4 The method of this embodiment will be described in detail.

[0029] I. Local Propagation Delay Estimation for Data-Physical Coordination.

[0030] All meteorological stations are divided into several regions according to their spatial location. Within each region, the propagation delay of meteorological observation data between any two stations within the region is estimated from the perspectives of meteorological observation data and physical laws. The data delay probability distribution and physical delay probability distribution are adaptively fused to obtain the local propagation delay between station pairs and their reliability weights.

[0031] Specifically: Given an irregular spatial distribution Each weather station, in the past Data was collected at that time. Each meteorological element. The spatial location of the station is represented as follows: That is, longitude, latitude, and altitude. The observation time is expressed as... This corresponds to the year, month, day, and hour of each observation time. Meteorological elements, i.e., meteorological observation data, are denoted as... ,in Indicates the first A site in the past The observation value at each time point.

[0032] To facilitate the model's comprehensive use of spatiotemporal information for prediction, the observed data and temporal and spatial information are embedded into a unified latent representation space, formally expressed as: (1); in, For the embedded representation of all sites, the three embedding functions are implemented by linear transformations that map the input to the hidden dimensions. .

[0033] Based on this, the future forecasts for each meteorological station are... Predictions are made based on meteorological observation data at specific times.

[0034] First, regarding the original Each meteorological station is clustered based on its spatial location to obtain several spatially adjacent regions. ,in Indicates the first Each region.

[0035] For any two stations in the same area Calculate candidate delay Cross-correlation score : (2); Used to measure the site observation sequence After time shift With the site observation sequence The degree of consistency between them.

[0036] To reduce computational complexity, fast Fourier transform and inverse Fourier transform are used to accelerate cross-correlation calculations: (3); in, and Indicates site and The observation sequence, and These represent the Fast Fourier Transform (FFT) and its inverse Fourier Transform (IFFT), respectively. This indicates the complex conjugate operation.

[0037] This step transforms the high-complexity computation of traditional cross-correlation into frequency domain multiplication, making it applicable to large-scale multi-site scenarios.

[0038] Subsequently, a propagation delay prior based on physical laws is constructed.

[0039] Set up a site With the site The spherical distance between them is The sampling interval is The propagation speed of meteorological elements along the direction of the station connection is a learnable parameter. The ideal propagation delay between sites is expressed as: (4).

[0040] by Construct candidate delays centered on [the target group]. Gaussian physical prior distribution: (5); in, Indicates propagation delay The probability that this value is in The nearest maximum, i.e., propagation delay estimation based on physical laws, tends to estimate the delay between stations as the time required for meteorological elements to propagate in the atmosphere.

[0041] To accommodate the propagation uncertainty of different site pairs, variance Taking into account both station distance and the degree of fluctuation in the observation sequence, it can be expressed as: (6); in, It represents the median absolute deviation of meteorological observation data and is used to measure the degree of local fluctuation in a time series. To balance the weights of propagation distance and data volatility, The basic bias term is used to avoid Too small.

[0042] Subsequently, to avoid the influence of spurious peaks at the boundary of cyclic correlation in finite-length sequences, the candidate delays were limited to a preset range. Inside, among which That is, only retain Corresponding and The remaining candidates are delayed and discarded directly.

[0043] Furthermore, to simultaneously express both early and late propagation directions, the candidate delay is expanded into a signed interval. .in, Indicates site Weather phenomena relative to In advance, that is, the direction of transmission is ; Indicates site Weather phenomena relative to Lag, meaning the direction of propagation is .

[0044] Building upon this, the definitions of data-driven cross-correlation scores and physical propagation delay priors are extended to the signed axis using the following rules: (7); The meaning of this rule is: site Compared to in advance The probability equals the site Compared to Delay The possibility of this ensures that the representations made earlier and later remain consistent in physical meaning.

[0045] After normalizing the data-driven cross-correlation score and the physical propagation delay prior, the data delay probability distribution is obtained. and physical delay probability distribution .

[0046] Finally, the data latency probability distribution and the physical latency probability distribution are adaptively fused. Based on The normalized entropy is used to calculate the confidence level of the data delay probability distribution. .

[0047] (8); in, and They represent Entropy and confidence. When When a clear, sharp peak is observed, its entropy value is low, indicating a strong discriminatory power between the probabilities of different candidate delays. The confidence level is relatively high; conversely, when the distribution is relatively flat or there are multiple competing peaks, the entropy is high and the confidence level is low, indicating that it is difficult to reliably determine the propagation delay by relying solely on observation data.

[0048] based on The fusion gating coefficient is defined as: (9); in, This represents the upper limit of the maximum proportion of the probability distribution of the propagation delay of physical laws in the fusion process. Both are hyperparameters used to adjust the sensitivity of the gating to changes in confidence.

[0049] The final fusion distribution is as follows: (10); And further by taking The maximum value obtained from the site and The propagation delay between them and their corresponding reliability weights.

[0050] The calculation method is as follows: (11); in, Indicates site and Propagation delay estimation between; This represents the probability value corresponding to the propagation delay, i.e., the reliability weight, used to characterize... Reliability.

[0051] II. Construction of Hierarchical Representation for Region Delay Awareness.

[0052] Propagation consistency constraints are constructed by leveraging local propagation delays. These constraints are then used to integrate the local propagation delays between multiple station pairs into a regional delay. Based on this regional delay, time misalignment corrections are applied to the weather representations of stations within the region. The corrected weather representations within the same region are then aggregated to obtain a higher-level regional representation. This process is repeated to progressively construct a hierarchical weather representation from the station level to the regional level and finally to the global level.

[0053] Specifically: Targeting the region Any site in China Local propagation delay and reliability weight For each station in the region Introducing regional delay This represents the relative arrival time of the station during the regional propagation process. Propagation consistency constraints are constructed using local propagation delays, ensuring... as close as possible And solve the following weighted least squares problem: (12); Among them, constraints Used to eliminate the non-uniqueness of global translation.

[0054] This optimization problem can be equivalently represented as a Laplace linear system on a weighted directed graph: nodes correspond to stations, edges correspond to the propagation delay of station pairs, and edge weights correspond to reliability weights. By solving this constrained linear system, a globally consistent propagation reference can be obtained within the region. And based on regional delay Correcting time discrepancies in weather reports from various stations within the region: (13); in, Indicates basis The weather representations for each station within the region are shifted along the time dimension to align the weather representations that were originally misaligned due to propagation delay. To maintain the sequence length, cyclic padding is used for portions exceeding the original time range.

[0055] After this step This indicates the weather forecast for stations within the region after regional propagation correction.

[0056] Subsequently, the weather representations of stations within the same area are averaged and aggregated to obtain the next layer of virtual station weather representations: (14); in, Indicates the region The average of the weather data from all stations within the region is calculated to obtain the result. It can be regarded as representing a region By performing the above operations on all regions within the current level, a virtual site containing all site data can be obtained as the representation set for the next level. .

[0057] Subsequently, these virtual nodes were used as new research objects, and the process of local delay estimation, regional delay recovery, temporal misalignment correction, and regional aggregation was repeatedly performed to gradually construct a hierarchical representation from sites and regions to the global level. .

[0058] Generally, the first Layer contains One site ( These sites were clustered into There are several regions, and each region, when aggregated, forms a site in the next level. Therefore, there are... .

[0059] III. Hierarchical spatiotemporal modeling.

[0060] Based on hierarchical weather representation, the spatial propagation relationships and temporal evolution patterns at different levels are modeled. Spatial modeling is used to depict the propagation direction and sequence of weather processes, while temporal modeling is used to characterize dynamic changes at different scales. Finally, the spatiotemporal characteristics of each level are integrated, and the prediction results of future meteorological elements are obtained through linear mapping.

[0061] Specifically: Based on hierarchical weather representation, spatial and temporal features are modeled at each level, and cross-level fusion is performed to obtain spatiotemporal features that take into account both local fluctuations and large-scale trends.

[0062] In the spatial dimension, for each region within each level, based on region-level delay... The stations within the region are sorted to explicitly reconstruct the propagation sequence of weather processes in the region; then, causal convolution is used to extract spatial features from the sorted node sequence. (15); in, Indicates the region Spatial characteristics of all stations within the area, This indicates that stations within a region are sorted based on regional latency. This represents causal convolution. Causal convolution only aggregates information from nodes before the current position. Therefore, its feature transfer process naturally follows the propagation order, making the resulting spatial representation more consistent with the actual propagation structure of weather processes.

[0063] Furthermore, the first By summing up all spatial features below the first layer, we obtain the first... Spatial characteristics of layers .

[0064] In the temporal dimension, different levels of weather representation correspond to weather processes within different spatial ranges, and their temporal variation rhythms typically differ significantly. Lower-level representations reflect more localized rapid fluctuations and short-term disturbances, while higher-level representations tend to focus on the slow evolution and overall trends of large-scale weather systems. Therefore, a layer-dependent temporal modeling strategy is introduced to enhance the model's ability to represent different temporal variation patterns.

[0065] For the Observation data of any station in the layer Perform a discrete wavelet transform to decompose it into a low-frequency approximate component and multiple high-frequency detail components: (16); in, Indicates after the first The approximate coefficients obtained after the decomposition. Indicates the first The detail coefficients obtained from the decomposition. Indicates the first The wavelet decomposition order corresponding to the layer The number of levels increases, allowing lower-level representations to retain more local rapid fluctuations, while higher-level representations can enhance the characterization of large-scale slow evolutionary trends.

[0066] Assuming there is a total Each level The calculation formula is: (17); in and represents the minimum and maximum allowed number of decompositions, respectively; both are hyperparameters.

[0067] After obtaining components in different frequency bands, learnable filters are used to model the low-frequency approximate components and the high-frequency detail components at each level, and the time series is reconstructed by inverse wavelet transform. (18); in, , Represent the effects acting on the low-frequency approximate component and the first... The learnable filters for high-frequency detail components are all constructed from two layers of linear mappings and an intermediate GELU activation function. Inverse wavelet transform operation; Indicates the first The first in the layer The time features of each site are stacked to obtain the time features of all sites. Temporal characteristics of layers .

[0068] At each level, the time features obtained from the modeling will be... Spatial features The features are then concatenated; following the correspondence from the hierarchical aggregation, high-level features are extended back to the low-level site dimensions, and adaptive weights are used to integrate the features from each level. (19); in, Represents a unified hierarchical spatiotemporal characteristic. This indicates a splicing operation. Indicates the first Adaptive weights for layer features.

[0069] Ultimately, in order to predict the future Data at each time point is mapped using a simple linear mapping to... : .

[0070] IV. Explanation of Evaluation Indicators

[0071] Weather data is a typical type of time series data, and predicting it is a regression problem, so we use evaluation metrics for regression problems.

[0072] ; (20).

[0073] Among them, mean squared error (MSE) is the mean of the squares of the differences between the predicted value and the true value. The smaller the mean squared error, the closer the prediction result is to the true value. Mean absolute error (MAE) is the mean of the absolute values ​​of the differences between the predicted value and the true value. The smaller the mean absolute error, the closer the prediction result is to the true value. Indicates the first Each weather station at time The actual value, Indicates the first Each weather station at time The predicted value, It is the total number of weather stations. It represents the time frame predicted by each weather station.

[0074] Example 2 This embodiment provides a delayed-sensing hierarchical spatiotemporal weather forecasting system, including: The segmentation module is configured to construct a weather representation for each meteorological station based on its spatial location, meteorological observation data, and corresponding observation time, and to divide all stations into multiple regions according to their spatial location. The delay estimation module is configured to calculate the cross-correlation score under candidate delay based on meteorological observation data for any two stations in the same area, and to construct a propagation delay prior based on physical laws. In this way, the data delay probability distribution and the physical delay probability distribution are obtained respectively, and after fusion, the local propagation delay between the station pairs and their weights are obtained. The hierarchical modeling module is configured to obtain the regional delay based on the local propagation delay and its weight between the station pairs, perform time misalignment correction on the weather representation of the stations in the region based on the regional delay, aggregate the corrected weather representation of the stations in the same region, and use it as the virtual weather representation of the next layer, until there is only one virtual weather representation in the current layer, thereby obtaining the hierarchical weather representation. The spatiotemporal modeling module is configured to extract spatial features from stations within the region after sorting them by regional delay for each layer in the hierarchical weather representation, extract temporal features after frequency band decomposition and reconstruction of meteorological observation data from each station, and fuse the spatial and temporal features of each layer to obtain the prediction results of meteorological observation data.

[0075] It should be noted that the above modules correspond to the steps described in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0076] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0077] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0078] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0079] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0080] The method in Example 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0081] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0082] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0083] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0084] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0085] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0086] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A hierarchical spatiotemporal weather prediction method with delayed sensing, characterized in that, include: Based on the spatial location of each meteorological station, meteorological observation data, and the corresponding observation time, a weather representation of the station is constructed, and all stations are divided into multiple regions according to their spatial location. For any two stations in the same area, the cross-correlation score under the candidate delay is calculated based on meteorological observation data, and a propagation delay prior based on physical laws is constructed. The data delay probability distribution and physical delay probability distribution are obtained respectively and then fused to obtain the local propagation delay and its weight between the station pairs. The regional delay is obtained based on the local propagation delay and its weight between the stations. The time misalignment of the weather representations of the stations in the region is corrected based on the regional delay. The corrected weather representations of the stations in the same region are aggregated and used as the virtual weather representations of the next layer until there is only one virtual weather representation in the current layer. This process is repeated to obtain a hierarchical weather representation. For each layer in the hierarchical weather representation, spatial features are extracted after sorting the stations in the region according to regional delay. Temporal features are extracted after frequency band decomposition and reconstruction of the meteorological observation data of each station. The spatial and temporal features of each layer are then fused to obtain the prediction results of the meteorological observation data.

2. The delayed-sensing hierarchical spatiotemporal weather prediction method as described in claim 1, characterized in that, The process of constructing a propagation delay prior based on physical laws includes: Calculate the ideal propagation delay between site pairs: ； With ideal propagation delay Construct candidate delays centered on [the target group]. Gaussian physical prior distribution: ； in, For the site With the site The spherical distance between them The sampling interval is... The propagation speed of meteorological observation data along the direction of the line connecting the stations; For the propagation delay prior, let the propagation delay be represented. The possibility; Let Variance be the variance.

3. The delayed-sensing hierarchical spatiotemporal weather prediction method as described in claim 1, characterized in that, The data latency probability distribution and the physical latency probability distribution are as follows: Candidate delay Limited to a preset range Inside, only retain The corresponding cross-correlation score and , , This represents the total time of meteorological observation data collection; Expand the candidate delay to a signed interval. Extend the cross-correlation score and propagation delay prior to the signed axis: ； in, Indicates site Weather phenomena relative to In advance, that is, the direction of transmission is ; Indicates site Weather phenomena relative to Lag, meaning the direction of propagation is ; After normalizing the cross-correlation score and the propagation delay prior, the data delay probability distribution is obtained. and physical delay probability distribution .

4. The delayed-sensing hierarchical spatiotemporal weather prediction method as described in claim 1, characterized in that, The process of obtaining the local propagation delay and its weight between site pairs includes: Based on data latency probability distribution Calculate the confidence level of normalized entropy ; based on Calculate the fusion gating coefficient: ;in, This represents the upper limit of the maximum proportion of the physical delay probability distribution in the fusion. Both are hyperparameters used to adjust the sensitivity of the fusion gating coefficients to changes in confidence level. Based on the fusion gating coefficient The data delay probability distribution and the physical delay probability distribution are fused to obtain a fused distribution. The maximum value of the fused distribution is taken, and the delay that produces the maximum value is taken as the local propagation delay, with the maximum value as the weight.

5. The delayed-sensing hierarchical spatiotemporal weather prediction method as described in claim 1, characterized in that, Based on the local propagation delay between site pairs and their weights The process of obtaining regional latency includes: for each station within the region Introducing regional delay And construct propagation consistency constraints to make Approaching Solve the following weighted least squares problem to obtain the region-level delay. : ； Among them, constraints Used to eliminate the non-uniqueness of global translation.

6. The delayed-sensing hierarchical spatiotemporal weather prediction method as described in claim 1, characterized in that, The process of extracting time features includes: For the first Sites in the layer The observed data were subjected to wavelet decomposition, and the different frequency band components were modeled separately and the time series was reconstructed using inverse wavelet transform; among them, wavelet decomposition The number of times is: ； in, and These represent the minimum and maximum allowed number of decompositions, respectively; This represents the total number of floors.

7. A delayed-sensing hierarchical spatiotemporal weather forecasting system, characterized in that, include: The segmentation module is configured to construct a weather representation for each meteorological station based on its spatial location, meteorological observation data, and corresponding observation time, and to divide all stations into multiple regions according to their spatial location. The delay estimation module is configured to calculate the cross-correlation score under candidate delay based on meteorological observation data for any two stations in the same area, and to construct a propagation delay prior based on physical laws. In this way, the data delay probability distribution and the physical delay probability distribution are obtained respectively, and after fusion, the local propagation delay between the station pairs and their weights are obtained. The hierarchical modeling module is configured to obtain the regional delay based on the local propagation delay and its weight between the station pairs, perform time misalignment correction on the weather representation of the stations in the region based on the regional delay, aggregate the corrected weather representation of the stations in the same region, and use it as the virtual weather representation of the next layer, until there is only one virtual weather representation in the current layer, thereby obtaining the hierarchical weather representation. The spatiotemporal modeling module is configured to extract spatial features from stations within the region after sorting them by regional delay for each layer in the hierarchical weather representation, extract temporal features after frequency band decomposition and reconstruction of meteorological observation data from each station, and fuse the spatial and temporal features of each layer to obtain the prediction results of meteorological observation data.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images