Short-time heavy rainfall area collaborative early warning method and system based on multi-source fusion data

Abstract

The invention relates to a short-time heavy rainfall region collaborative early warning method based on multi-source fusion data. The method comprises the following steps: firstly, acquiring multi-source heterogeneous data of a ground rainfall station, a radar, a satellite and the like, and converting the multi-source heterogeneous data into a rainfall monitoring data set by adopting multi-scale interpolation and standardization processing; secondly, preprocessing the data set to extract precipitation space features, and combining wind field and air pressure field data to input a time sequence prediction model to generate a precipitation path prediction sequence; and then, processing the prediction sequence by using a graph structure segmentation algorithm, and generating a high-precision early warning region boundary sequence. And finally, carrying out space overlapping degree calculation on the early warning region boundary sequence and a historical disaster region, and when a preset threshold value is exceeded, comprehensively analyzing parameters such as a disaster type and an influence range by adopting a probability model to generate a graded early warning sequence. By adopting the method, the integrity and accuracy of the rainfall monitoring data set can be improved, and the timeliness and reliability of short-time heavy rainfall early warning can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of meteorological monitoring and early warning technology, and in particular relates to a method and system for collaborative early warning of short-term heavy rainfall areas based on multi-source fusion data. Background Art

[0002] With the advancement of meteorological monitoring and early warning technologies, collaborative regional early warning technologies for short-term heavy precipitation based on multi-source fusion data have emerged. Short-term heavy precipitation, characterized by its short duration and high intensity, can easily trigger secondary disasters such as urban waterlogging and flash floods, posing a serious threat to socioeconomic development and public safety. Existing early warning methods, data from multiple sources—such as ground rain gauges, weather radar, and satellites—disparate in temporal and spatial resolution, data structure, and acquisition frequency, leading to information inconsistencies during data fusion. Traditional interpolation and normalization methods struggle to dynamically optimize parameters based on data source characteristics, compromising the integrity and accuracy of precipitation monitoring data. In the precipitation path prediction phase, existing time-series prediction models inadequately model the coupled relationship between wind and pressure fields and spatial characteristics of precipitation, lacking a quantitative mechanism to effectively characterize the dynamic impact of wind and pressure fields. Regarding warning area generation, graph-based segmentation algorithms fail to fully integrate geographic information such as road network topology and terrain slope, resulting in limited accuracy in delineating warning area boundaries. Existing probabilistic models, during the warning classification process, lack dynamic weighting strategies for assessing the overlap between historical disaster areas and the current warning area, reducing the reliability of warning results. Summary of the Invention

[0003] Based on this, it is necessary to provide a short-term heavy rainfall regional collaborative warning method and system based on multi-source fusion data, which can effectively improve the timeliness and reliability of short-term heavy rainfall warnings and provide accurate data support for disaster prevention and emergency decision-making, in response to the above technical problems.

[0004] First, this application provides a method for regional collaborative early warning of short-term heavy rainfall based on multi-source fusion data, including:

[0005] Obtain data from ground rain gauges, radars, and satellites; apply multi-scale interpolation and normalization to the data to obtain a precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity.

[0006] The precipitation monitoring dataset is preprocessed and the spatial characteristics of precipitation are extracted. The wind field and pressure field data are combined to input the time series prediction model to generate the precipitation path prediction sequence.

[0007] The graph structure segmentation algorithm is used to process the precipitation path prediction sequence, integrating road network and slope characteristics to generate a high-precision warning area boundary sequence.

[0008] If the overlap between the warning area boundary sequence and the historical disaster area exceeds a preset threshold, a probabilistic model is used to generate a graded warning sequence.

[0009] In one embodiment, multi-scale interpolation and normalization are performed on the data to obtain a precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity, including:

[0010] Obtain the temporal resolution of ground rain gauge data, the spatial coverage of radar data, and the band characteristics of satellite data; ground rain gauge data contains precipitation records at discrete points.

[0011] Generate sliding window parameters for multi-scale interpolation based on the time resolution; the sliding window parameters are used to control data alignment in the time dimension.

[0012] The grid density threshold of the interpolation algorithm is determined based on the spatial coverage; the grid density threshold is negatively correlated with the radar scanning radius.

[0013] The data are processed using a multi-scale interpolation algorithm based on the sliding window parameters and the grid density threshold to obtain an interpolated data set.

[0014] The water vapor absorption index is calculated according to the band characteristics, and the weight factor for normalization is obtained.

[0015] The interpolation dataset is fused according to the weight factor to obtain a three-dimensional matrix containing timestamps, longitude and latitude, and precipitation intensity.

[0016] The three-dimensional matrix is projected and transformed with the preset coordinate system to obtain the precipitation monitoring dataset.

[0017] In one embodiment, after obtaining the precipitation monitoring dataset, the method further includes:

[0018] The abnormal precipitation area is divided according to the grid density threshold, and the interpolation algorithm is dynamically iteratively updated to obtain the dynamically updated data.

[0019] The sliding window parameters are used to smooth the time series of the dynamically updated data to obtain a smoothed dataset covering the missing period.

[0020] The normalization coefficient of the normalization process is adjusted according to the water vapor absorption index; the normalization coefficient is used for the intensity unit conversion between different data sources.

[0021] The projected data are superimposed with the data of the smoothed dataset to obtain the final precipitation monitoring dataset.

[0022] In one embodiment, the precipitation monitoring dataset is preprocessed and spatial features of precipitation are extracted, and wind field and pressure field data are combined to input into a time series prediction model to generate a precipitation path prediction sequence, including:

[0023] Get the original precipitation intensity distribution in the precipitation monitoring dataset; the original precipitation intensity distribution contains spatial coordinates and timestamp information.

[0024] The original precipitation intensity distribution is mapped into a spatiotemporal interpolation grid to obtain an interpolated precipitation intensity matrix; the spatiotemporal interpolation grid covers the geographical range of the target area.

[0025] Obtain the vector direction field from the wind field data and the gradient field from the pressure field data.

[0026] The dynamic field coupling coefficient is calculated based on the vector direction field and the gradient field; the dynamic field coupling coefficient reflects the association weight of the wind pressure field on the spatial characteristics of precipitation.

[0027] The spatial gradient features in the precipitation intensity matrix are fused with the dynamic field coupling coefficient to generate a spatiotemporal feature fusion tensor; the spatial gradient features include the intensity change rate of adjacent grid cells.

[0028] The spatiotemporal feature fusion tensor is input into the time series prediction model to output the probability distribution of precipitation paths for multiple time steps in the future; the probability distribution of precipitation paths contains a sequence of spatial displacement vectors.

[0029] A precipitation path prediction sequence is generated based on the spatial displacement vector sequence; the precipitation path prediction sequence marks the movement trajectory of the precipitation core area in the target area.

[0030] In one embodiment, the spatiotemporal feature fusion tensor is calculated using the following formula:

[0031]

[0032] Among them, T represents the spatiotemporal feature fusion tensor, I represents the spatiotemporal precipitation intensity matrix, G represents the spatial gradient feature matrix, Λ represents the dynamic field coupling coefficient, Attention(·) represents the attention mechanism function, and W α represents the learnable weight matrix, represents the scaling factor, σ represents the GELU activation function, and b represents the bias vector.

[0033] In one embodiment, a graph structure segmentation algorithm is used to process a precipitation path prediction sequence, integrating road network and slope features to generate a high-precision warning area boundary sequence, including:

[0034] The graph structure segmentation algorithm is used to process the precipitation path prediction sequence and generate a path graph structure containing node connection relationships.

[0035] Obtain topological connectivity data and slope characteristic data of the road network; the slope characteristic data includes the slope change rate of each grid cell.

[0036] The neighborhood weight of each node in the path graph structure is determined based on the topological connectivity data, and the path similarity matrix is calculated by combining the slope change rate.

[0037] The path similarity matrix is processed using a spatial smoothing algorithm to generate a grid risk value distribution containing fusion features.

[0038] The dynamic threshold of the grid risk value is set based on the path similarity matrix and the slope change rate weight.

[0039] The grid risk value distribution is binarized based on the dynamic threshold to obtain the initial warning area boundary.

[0040] The initial warning area boundary is optimized by iteratively adjusting the neighborhood weights and smoothing parameters to obtain a high-precision warning area boundary sequence containing a time series.

[0041] In one embodiment, the method further comprises:

[0042] The warning area is spatially grid-coded according to the spatial coordinate data of the warning area boundary sequence to obtain a gridded boundary sequence.

[0043] A disaster history distribution database that intersects with the gridded boundary sequence is extracted from the disaster history database; the disaster history distribution database includes disaster type and impact range parameters.

[0044] Calculate the regional overlap index between the gridded boundary sequence and the historical disaster distribution database; the regional overlap index includes spatial coverage and temporal frequency factors.

[0045] When the regional overlap index exceeds the preset threshold, the dynamic weight parameter adjustment algorithm is used to process the disaster type and impact range parameters to obtain the adjusted dynamic weight parameters.

[0046] The dynamic weight parameters are input into the probability model, and a graded warning sequence containing warning levels and confidence values is output.

[0047] The hierarchical warning sequence is input into the streaming computing framework to obtain a structured warning message dataset; the structured warning message dataset includes the warning level and the associated data source identifier.

[0048] According to the warning level, the preset anomaly detection model is matched to generate a set of abnormal events with risk scores.

[0049] The abnormal event set is input into the priority queue of the distributed messaging system to generate a real-time warning push sequence with terminal identification and push time; the priority queue dynamically adjusts the message sorting based on the risk score.

[0050] According to the associated data source identifier, the preset feedback receiving interface is matched, the parameter weights of the anomaly detection model are updated to process the newly added warning message data set, and the updated anomaly event set is obtained and added to the priority queue.

[0051] Secondly, this application also provides a short-term heavy rainfall regional collaborative early warning system based on multi-source fusion data, the system includes:

[0052] The data processing module is used to obtain data from ground rain gauges, radars, and satellites. Multi-scale interpolation and standardization are performed on the data to obtain a precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity.

[0053] The precipitation path prediction module is used to preprocess the precipitation monitoring data set and extract the precipitation spatial characteristics, and combine the wind field and pressure field data to input the time series prediction model to generate the precipitation path prediction sequence.

[0054] The warning grade generation module is used to process the precipitation path prediction sequence using a graph structure segmentation algorithm, integrate road network and slope characteristics, and generate a high-precision warning area boundary sequence; if the overlap between the warning area boundary sequence and the historical disaster area exceeds a preset threshold, a probability model is used to generate a graded warning sequence.

[0055] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0056] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the above method when executed by a processor.

[0057] The aforementioned collaborative regional early warning method, system, computer device, and storage medium for short-term heavy precipitation based on multi-source fusion data first acquires heterogeneous data from multiple sources, including ground rain gauges, radar, and satellite data. Multi-scale interpolation and normalization are used to address differences in temporal resolution, spatial coverage, and data structure, transforming the data into a unified precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity. Second, the precipitation monitoring dataset is preprocessed to extract spatial characteristics of precipitation. The dynamic field coupling coefficient is calculated by combining wind vector direction and pressure gradient data to quantify the impact of wind pressure on precipitation. This coefficient is then fused with the precipitation intensity matrix and fed into a time series prediction model to generate a precipitation path prediction sequence. The prediction sequence is then processed using a graph segmentation algorithm, integrating road network topology and terrain slope characteristics. Node neighborhood weights and a path similarity matrix are calculated, and then segmented using a spatial smoothing algorithm and dynamic thresholding to generate a high-precision warning area boundary sequence. Finally, the spatial overlap between the warning area boundary sequence and historical disaster areas is calculated. When the overlap exceeds a preset threshold, a probabilistic model is used to comprehensively analyze parameters such as disaster type and impact range to generate a graded warning sequence. This method can improve the integrity and accuracy of precipitation monitoring datasets, enhance the ability to capture the evolution of precipitation paths, effectively improve the timeliness and reliability of short-term heavy rainfall warnings, and provide accurate data support for disaster prevention and emergency decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] Figure 1 A flowchart of a method for regional collaborative early warning of short-term heavy rainfall based on multi-source fusion data provided by an embodiment of the present invention;

[0060] Figure 2 This is a structural block diagram of a short-term heavy rainfall regional collaborative early warning system based on multi-source fusion data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0062] In one embodiment, Figure 1As shown, this application provides a short-term heavy rainfall regional collaborative early warning method based on multi-source fusion data, including:

[0063] Step S101: Acquire ground rain gauge, radar, and satellite data; apply multi-scale interpolation and normalization to the data to obtain a precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity.

[0064] Specifically, precipitation data is obtained through ground-based rain gauges, radar, and satellites. Ground-based rain gauges provide records of precipitation intensity at discrete points, radar data provides scanning information with high temporal and spatial resolution, and satellite data covers large areas and includes multi-band spectral characteristics. Due to differences in temporal resolution, spatial coverage, and data structure among various data sources, a multi-scale interpolation algorithm is used to dynamically adjust the interpolation grid density and time window parameters based on data characteristics. Standardized processing methods are then used to normalize and unify the data format, ultimately forming a structured precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity.

[0065] Step S102 , preprocessing the precipitation monitoring data set and extracting precipitation spatial features, combining the wind field and pressure field data to input the time series prediction model to generate a precipitation path prediction sequence.

[0066] Specifically, the precipitation monitoring dataset is preprocessed through outlier detection and missing value filling. The spatial gradient characteristics of precipitation are extracted by calculating the intensity change rate of adjacent grid cells. The vector direction field from the wind field data and the gradient field from the pressure field data are simultaneously acquired. A coupling model is constructed to calculate the dynamic field coupling coefficient and quantify the weight of the wind pressure field's influence on the spatial characteristics of precipitation. The precipitation intensity matrix is fused with the weighted spatial gradient characteristics to generate a spatiotemporal feature fusion tensor. This tensor is then fed into a deep learning-based time series prediction model. Through training on historical data, the probability distribution of precipitation paths for multiple future time steps is output, thereby generating a predicted sequence of movement trajectories for the precipitation core area.

[0067] Step S103 : Processing the precipitation path prediction sequence using a graph structure segmentation algorithm, integrating road network and slope features, and generating a high-precision warning area boundary sequence.

[0068] A graph-structured segmentation algorithm was used to perform topological analysis on the precipitation path prediction sequence, converting the predicted trajectories into a graph structure containing node connectivity. Topological connectivity data of the road network and terrain slope change rate were incorporated, and the correlation between precipitation risks in different regions was quantified by calculating node neighborhood weights and a path similarity matrix. A spatial smoothing algorithm was used to optimize the risk value distribution. Dynamic thresholds were set based on precipitation intensity and geographic characteristics, and the grid risk value distribution was binarized to form the initial warning area boundaries. The boundaries were optimized by iteratively adjusting the algorithm parameters, ultimately generating a high-precision warning area boundary sequence containing time series information.

[0069] Step S104: If the overlap between the warning area boundary sequence and the historical disaster area exceeds a preset threshold, a probability model is used to generate a graded warning sequence.

[0070] Specifically, the generated warning region boundary sequence is spatially overlaid with a database of historical disaster regions to calculate a regional overlap index, which incorporates spatial coverage and temporal frequency factors. When the overlap exceeds a preset threshold, a probabilistic model is used to analyze parameters such as disaster type and historical impact range. A dynamic weight adjustment algorithm is used to optimize the influence of each parameter, and a graded warning sequence containing warning levels and confidence values is output, providing a decision-making basis for disaster warning and emergency response.

[0071] The proposed method for regional coordinated early warning of short-term heavy precipitation based on multi-source fusion data first acquires heterogeneous data from multiple sources, including ground rain gauges, radar, and satellite data. Multi-scale interpolation and normalization are used to address differences in temporal resolution, spatial coverage, and data structure, transforming the data into a unified precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity. Second, the precipitation monitoring dataset is preprocessed to extract spatial characteristics of precipitation. The dynamic field coupling coefficient is calculated by combining wind vector direction and pressure gradient data to quantify the impact of wind pressure on precipitation. This coefficient is then fused with the precipitation intensity matrix and fed into a time series prediction model to generate a precipitation path prediction sequence. The prediction sequence is then processed using a graph segmentation algorithm, integrating road network topology and terrain slope characteristics. Node neighborhood weights and path similarity matrices are calculated, and segmentation is performed using a spatial smoothing algorithm and dynamic thresholding to generate a high-precision warning area boundary sequence. Finally, the spatial overlap between the warning area boundary sequence and historical disaster areas is calculated. When the overlap exceeds a preset threshold, a probabilistic model is used to comprehensively analyze parameters such as disaster type and impact range to generate a graded warning sequence. This method can improve the integrity and accuracy of precipitation monitoring datasets, enhance the ability to capture the evolution of precipitation paths, effectively improve the timeliness and reliability of short-term heavy rainfall warnings, and provide accurate data support for disaster prevention and emergency decision-making.

[0072] In one embodiment, multi-scale interpolation and normalization are performed on the data to obtain a precipitation monitoring dataset including timestamps, longitude and latitude, and precipitation intensity, which may include the following steps:

[0073] Step S201, obtaining the temporal resolution of ground rain gauge data, the spatial coverage of radar data, and the band characteristics of satellite data; the ground rain gauge data includes precipitation records at discrete points.

[0074] Step S202 : Generate sliding window parameters for multi-scale interpolation according to the time resolution; the sliding window parameters are used to control data alignment in the time dimension.

[0075] Step S203: determining a grid density threshold of the interpolation algorithm based on the spatial coverage range; the grid density threshold is negatively correlated with the radar scanning radius.

[0076] Step S204 : Processing the data using a multi-scale interpolation algorithm based on the sliding window parameters and the grid density threshold to obtain an interpolated data set.

[0077] Step S205 , calculating the water vapor absorption index according to the band characteristics to obtain a weight factor for normalization processing.

[0078] Step S206 , performing a fusion operation on the interpolation data set according to the weight factor to obtain a three-dimensional matrix including timestamps, longitude and latitude, and precipitation intensity.

[0079] Step S207 , performing a projection transformation on the three-dimensional matrix and the preset coordinate system to obtain a precipitation monitoring data set.

[0080] Specifically, the temporal resolution of ground-based rain gauge data, the spatial coverage of radar data, and the band characteristics of satellite data are first determined. Ground-based rain gauges record precipitation information at discrete locations. Based on the temporal resolution, sliding window parameters for multiscale interpolation are generated to achieve data alignment in the temporal dimension. Based on the spatial coverage of the radar data, a grid density threshold, negatively correlated with the scanning radius, is determined to optimize spatial interpolation accuracy. Combining the sliding window parameters and the grid density threshold, a multiscale interpolation algorithm is used to process the multi-source data to form an interpolated dataset. Furthermore, by analyzing the band characteristics of the satellite data and calculating the water vapor absorption index, the weighting factors required for standardization are derived. Based on these factors, the interpolated dataset is fused to produce a three-dimensional matrix containing timestamps, longitude and latitude, and precipitation intensity. Finally, the three-dimensional matrix is projected onto a pre-set coordinate system to generate a standardized precipitation monitoring dataset.

[0081] This embodiment constructs a parameter-driven multi-scale interpolation and standardization system by quantifying the core parameters of each data source (temporal resolution, spatial coverage, and band characteristics). The dynamic setting of sliding window parameters and grid density thresholds solves the problems of alignment and resolution differences of multi-source data in the temporal and spatial dimensions, and effectively improves the accuracy of data interpolation; the weight factor calculation based on the water vapor absorption index realizes the scientific weighted fusion of different data sources, avoiding information bias in the data fusion process. The projection transformation operation ensures that the data is unified in the standard coordinate system, providing a standardized data basis for subsequent analysis. This method systematically improves the integrity, consistency, and availability of precipitation monitoring data, provides high-quality data support for precipitation feature analysis, path prediction, and warning model construction, and enhances the data processing efficiency of the short-term heavy rainfall warning system.

[0082] In one embodiment, after obtaining the precipitation monitoring dataset, the following steps may also be included:

[0083] Step S301 : dividing abnormal precipitation areas according to a grid density threshold and dynamically iterating and updating the interpolation algorithm to obtain dynamically updated data.

[0084] Step S302 : Time series smoothing is performed on the dynamically updated data using sliding window parameters to obtain a smoothed data set covering the missing period.

[0085] Step S303 , adjusting the normalization coefficient of the normalization process according to the water vapor absorption index; the normalization coefficient acts on the intensity unit conversion of different data sources.

[0086] Step S304 : superimposing the projected data with the smoothed data set to obtain a final precipitation monitoring data set.

[0087] Specifically, areas of abnormal precipitation are identified based on a defined grid density threshold. The interpolation algorithm is dynamically and iteratively updated for these areas. Interpolation parameters and computational logic are adjusted to optimize data accuracy and obtain dynamically updated data. Time series smoothing is then applied to the dynamically updated data using the sliding window parameters generated earlier. A filtering algorithm is then used to fill in missing data periods, forming a continuous smoothed dataset. The normalization coefficient used in the standardization process is adjusted based on the water vapor absorption index of the satellite data. This coefficient is used to unify the unit standards for precipitation intensity across different data sources. Finally, the projected data are overlaid and integrated with the smoothed dataset to eliminate data inconsistencies and redundancies, generating a final precipitation monitoring dataset with complete spatiotemporal information and unified unit standards.

[0088] This embodiment significantly improves the quality of precipitation monitoring data through multi-step collaborative optimization. The identification of abnormal regions based on grid density thresholds and the dynamic update of the interpolation algorithm effectively correct interpolation errors in complex precipitation scenarios and enhance data accuracy. Time series smoothing utilizes sliding window parameters to fill data gaps and ensure the continuity and integrity of temporal data. The final data overlay operation achieves a deep integration of spatial coordinate standardized data and time series optimized data, providing a high-quality data foundation for subsequent precipitation feature analysis, path prediction, and early warning model construction, thereby improving the data reliability and analytical effectiveness of the short-term heavy precipitation early warning system.

[0089] In one embodiment, preprocessing the precipitation monitoring dataset and extracting precipitation spatial features, combining wind field and pressure field data to input a time series prediction model to generate a precipitation path prediction sequence, may include the following steps:

[0090] Step S401: obtaining the original precipitation intensity distribution in the precipitation monitoring data set; the original precipitation intensity distribution includes spatial coordinates and timestamp information.

[0091] Step S402 : Mapping the original precipitation intensity distribution to a spatiotemporal interpolation grid to obtain an interpolated precipitation intensity matrix; the spatiotemporal interpolation grid covers the geographical range of the target area.

[0092] Step S403: Acquire the vector direction field in the wind field data and the gradient field in the pressure field data.

[0093] Step S404 , calculating a dynamic field coupling coefficient based on the vector direction field and the gradient field; the dynamic field coupling coefficient reflects the association weight of the wind pressure field with the spatial characteristics of precipitation.

[0094] Step S405 , performing matrix fusion on the spatial gradient features in the precipitation intensity matrix and the dynamic field coupling coefficient to generate a spatiotemporal feature fusion tensor; the spatial gradient features include the intensity change rate of adjacent grid cells.

[0095] Preferably, the spatial gradient characteristics in the precipitation intensity matrix are first extracted, and the structural differences in the spatial distribution of precipitation are quantified by calculating the precipitation intensity change rates of adjacent grid cells in the horizontal and vertical directions. At the same time, based on the dynamic field coupling coefficient calculated from the wind field vector direction field and the pressure field gradient field, the weight of the wind pressure field's influence on the spatial characteristics of precipitation is quantified. On this basis, matrix operations are used to fuse the spatial gradient characteristics with the dynamic field coupling coefficient: the dynamic field coupling coefficient is used as the weight, and the spatial gradient feature matrix is weighted. Through matrix dot multiplication or tensor product operations, the influence of the wind pressure field on the spatial variation of precipitation is integrated into the gradient characteristics, and finally a spatiotemporal feature fusion tensor containing spatiotemporal information and dynamic driving factors is generated. This tensor integrates the spatial variation law of precipitation intensity and the weight of the wind pressure field.

[0096] Step S406: Input the spatiotemporal feature fusion tensor into the time series prediction model to output the probability distribution of precipitation paths for multiple future time steps; the probability distribution of precipitation paths includes a sequence of spatial displacement vectors.

[0097] Step S407 , generating a precipitation path prediction sequence according to the spatial displacement vector sequence; the precipitation path prediction sequence marks the movement trajectory of the precipitation core area in the target area.

[0098] Furthermore, the original precipitation intensity distribution, which carries spatial coordinates and timestamp information, is first extracted from the precipitation monitoring dataset. By mapping the original precipitation intensity distribution onto a spatiotemporal interpolation grid covering the target area, an interpolated precipitation intensity matrix is constructed. The vector direction field from the wind field data and the gradient field from the pressure field data are simultaneously acquired. Based on this wind and pressure field data, a specific algorithm is used to calculate the dynamic field coupling coefficient, which quantifies the influence of the wind pressure field on the spatial characteristics of precipitation. The spatial gradient feature of the precipitation intensity matrix, namely the rate of change of intensity between adjacent grid cells, is further extracted and matrix-fused with the dynamic field coupling coefficient to form a spatiotemporal feature fusion tensor that incorporates spatiotemporal information and the influence of wind pressure. This tensor is then fed into a pre-trained time series prediction model. The model then calculates and outputs a probability distribution of precipitation paths for multiple future time steps, including a sequence of spatial displacement vectors. This sequence is then used to generate a precipitation path prediction sequence that identifies the movement trajectory of the precipitation core area.

[0099] This embodiment effectively improves the accuracy and reliability of precipitation path prediction through systematic data processing and feature fusion strategies. The precipitation intensity matrix is constructed based on the spatiotemporal interpolation grid to achieve the structuring and standardization of precipitation data. The introduction of the dynamic field coupling coefficient quantifies the impact of the wind pressure field on precipitation evolution and enhances the model's ability to characterize meteorological dynamic mechanisms; the matrix fusion of spatial gradient features and coupling coefficients enables the spatiotemporal feature fusion tensor to simultaneously reflect the precipitation's own changing laws and environmental field driving factors. Combined with the time series prediction model, it makes full use of historical data to learn the precipitation evolution law and outputs a prediction result containing a probability distribution, which not only provides a measure of the uncertainty of the prediction, but also accurately marks the trajectory of the precipitation core area through a sequence of spatial displacement vectors.

[0100] In one embodiment, the spatiotemporal feature fusion tensor can be calculated using the following formula:

[0101]

[0102] Among them, T represents the spatiotemporal feature fusion tensor, I represents the spatiotemporal precipitation intensity matrix, G represents the spatial gradient feature matrix, Λ represents the dynamic field coupling coefficient, Attention(·) represents the attention mechanism function, and W α represents the learnable weight matrix, represents the scaling factor, σ represents the GELU activation function, and b represents the bias vector.

[0103] This embodiment significantly improves the effectiveness and representation capabilities of spatiotemporal feature fusion through the collaborative operation of multiple components. The attention mechanism function can adaptively focus on areas that have a key impact on precipitation evolution, dynamically adjust the weights of spatial gradient features and dynamic field coupling coefficients, and strengthen the expression of important features; the learnable weight matrix gives the model the ability to autonomously optimize feature combinations based on data characteristics, enhancing model adaptability. The introduction of scaling factors prevents calculation results from being too large or too small, ensuring data stability; the GELU activation function simulates the complex characteristics of atmospheric processes through nonlinear transformations, enabling the fusion tensor to capture the nonlinear relationship between precipitation and the environmental field. The bias vector increases the flexibility of the model and allows the baseline value of the fusion result to be adjusted. The spatiotemporal feature fusion tensor generated by this calculation method integrates the precipitation intensity distribution, spatial variation patterns, and wind pressure field driving factors, providing more representative input for subsequent time series prediction models, effectively improving the accuracy and reliability of precipitation path predictions, and providing more powerful data support for short-term heavy precipitation warnings.

[0104] In one embodiment, a graph structure segmentation algorithm is used to process a precipitation path prediction sequence, integrate road network and slope features, and generate a high-precision warning area boundary sequence, which may include the following steps:

[0105] Step S501 : Processing the precipitation path prediction sequence using a graph structure segmentation algorithm to generate a path graph structure including node connection relationships.

[0106] Step S502: Acquire topological connectivity data and slope characteristic data of the road network; the slope characteristic data includes the slope change rate of each grid cell.

[0107] Step S503: determine the neighborhood weight of each node in the path graph structure according to the topological connectivity data, and calculate the path similarity matrix in combination with the slope change rate.

[0108] Preferably, the degree of association between each node in the path graph structure in the traffic network is quantified based on the road network topological connectivity data, and the attributes such as connectivity and accessibility between nodes are converted into neighborhood weights. The higher the weight value, the greater the influence of the node on the road network. At the same time, the terrain slope change rate is introduced, which reflects the terrain undulation of each grid unit and is used to measure the degree to which precipitation diffusion is hindered by the terrain. On this basis, the node neighborhood weight and the slope change rate are integrated, and the similarity between the paths of each node in the path graph structure is calculated by constructing a similarity measurement function. This calculation process is based on node attributes and geographical features, and quantifies the influence of road network and terrain factors on precipitation paths, and finally generates a path similarity matrix, which characterizes the degree of association between different paths.

[0109] Step S504: Process the path similarity matrix using a spatial smoothing algorithm to generate a grid risk value distribution containing fusion features.

[0110] Preferably, a spatial smoothing algorithm is used to process the path similarity matrix to reduce data noise interference and enhance the stability and continuity of the grid risk value distribution. This operation is based on the degree of correlation of precipitation paths represented by the path similarity matrix. By performing local neighborhood weighted averaging or Gaussian filtering on each element in the matrix, the risk values of adjacent grids influence each other and transition smoothly. During the processing, the weight coefficient is set according to the spatial distance, and grids with closer distances are given higher weights to highlight the consistency characteristics of precipitation risks in local areas. After processing by the spatial smoothing algorithm, the generated grid risk value distribution effectively integrates characteristics such as road network topological connectivity and slope change rate to form a quantitative result that can reflect the potential distribution of regional precipitation risk, providing a reliable data basis for the subsequent dynamic threshold segmentation of the warning area boundary.

[0111] Step S505 : setting a dynamic threshold of the grid risk value based on the path similarity matrix and the slope change rate weighting.

[0112] Step S506 , performing binary segmentation on the grid risk value distribution based on the dynamic threshold to obtain the initial warning area boundary.

[0113] Step S507 , optimizing the initial warning area boundary by iteratively adjusting the neighborhood weights and smoothing parameters to obtain a high-precision warning area boundary sequence including a time series.

[0114] First, a graph segmentation algorithm is used to process the precipitation path prediction sequence, converting it into a path graph structure containing node connectivity relationships and constructing a topological model of the precipitation path. Topological connectivity data of the road network and terrain slope characteristics are simultaneously acquired. The slope characteristics reflect the slope change rate of each grid cell. Based on the road network topological connectivity data, the neighborhood weight of each node in the path graph structure is determined. A path similarity matrix is calculated in combination with the slope change rate to quantify the degree of similarity between precipitation paths in different regions. A spatial smoothing algorithm is used to optimize the path similarity matrix to generate a grid risk value distribution that integrates the road network and slope characteristics. Based on the path similarity matrix and the slope change rate, a dynamic threshold for the grid risk value is set using a weighted approach. The grid risk value distribution is then binarized based on this dynamic threshold to obtain the initial warning area boundaries. Finally, the initial warning area boundaries are refined by iteratively adjusting the node neighborhood weights and smoothing algorithm parameters, forming a high-precision warning area boundary sequence that incorporates time series information.

[0115] This warning area boundary generation method significantly improves the accuracy and reliability of warning area delineation through multi-source information fusion and algorithm optimization. The precipitation path prediction sequence is converted into a graph structure, providing a structured data foundation for subsequent analysis. The introduction of road network topological connectivity and slope characteristic data fully considers the impact of the geographical environment on precipitation disasters, making the warning area delineation more consistent with the actual disaster diffusion pattern. The calculation of neighborhood weights and path similarity matrices effectively quantifies the precipitation risk associations in different regions. The spatial smoothing algorithm and dynamic threshold setting reduce data noise interference and improve the stability of boundary delineation. The initial boundaries are refined through an iterative optimization mechanism to ensure that the warning area boundaries accurately reflect the core impact range of precipitation and its evolutionary trends over time. The resulting high-precision warning area boundary sequence provides a precise spatial basis for graded warnings for short-term heavy precipitation, helping to improve the targeted nature of disaster warnings and the effectiveness of emergency responses.

[0116] In one embodiment, the method may further include the following steps:

[0117] Step S601 : performing spatial grid coding on the warning area according to the spatial coordinate data of the warning area boundary sequence to obtain a gridded boundary sequence.

[0118] Step S602: extracting a disaster history distribution library that intersects with the gridded boundary sequence from the disaster history database; the disaster history distribution library includes disaster type and impact range parameters.

[0119] Step S603: Calculate the regional overlap index between the gridded boundary sequence and the disaster historical distribution database; the regional overlap index includes a spatial coverage factor and a temporal frequency factor.

[0120] Step S604: When the regional overlap index exceeds a preset threshold, a dynamic weight parameter adjustment algorithm is used to process the disaster type and impact range parameters to obtain an adjusted dynamic weight parameter.

[0121] Step S605: Input the dynamic weight parameters into the probability model, and output a graded warning sequence including warning levels and confidence values.

[0122] Step S606: Input the graded warning sequence into the stream computing framework to obtain a structured warning message data set; the structured warning message data set includes the warning level and the associated data source identifier.

[0123] Step S607: Match the preset anomaly detection model according to the warning level to generate an abnormal event set with risk score.

[0124] Step S608: Input the abnormal event set into the priority queue of the distributed messaging system to generate a real-time warning push sequence that carries the terminal identifier and push time limit; the priority queue dynamically adjusts the message order based on the risk score.

[0125] Step S609 , matching the preset feedback receiving interface according to the associated data source identifier, updating the parameter weights of the anomaly detection model to process the newly added warning message data set, and obtaining an updated abnormal event set to the priority queue.

[0126] Specifically, the warning area is first spatially grid-encoded based on the spatial coordinate data of the warning area boundary sequence, forming a gridded boundary sequence and achieving a structured representation of the warning area. Subsequently, a historical disaster distribution database is selected from the disaster history database to identify entities that spatially intersect with the gridded boundary sequence. This distribution database covers parameters such as disaster type and impact range. A regional overlap index is derived by calculating the spatial coverage and temporal frequency factor between the gridded boundary sequence and the historical disaster distribution database. When this index exceeds a preset threshold, a dynamic weight parameter adjustment algorithm is used to optimize the disaster type and impact range parameters to obtain the adjusted dynamic weight parameters. The dynamic weight parameters are then input into a probabilistic model to generate a hierarchical warning sequence containing warning levels and confidence values. The hierarchical warning sequence is then input into a streaming computing framework and converted into a structured warning message dataset containing warning levels and associated data source identifiers. A pre-set anomaly detection model is then applied to the warning levels to generate a set of anomaly events with risk scores. This set of anomaly events is then fed into the priority queue of the distributed messaging system. The message order is dynamically adjusted based on the risk scores, generating a real-time warning push sequence that carries terminal identifiers and push timeliness. Finally, by associating the data source identifier with the preset feedback receiving interface, the parameter weights of the anomaly detection model are updated using the feedback data, the newly added warning message data set is processed, and the updated anomaly event set is included in the priority queue again.

[0127] This embodiment significantly improves the scientificity and practicality of short-term heavy rainfall warnings. The spatial grid coding and the correlation analysis of historical disaster data enable the warning level division to fully combine historical experience and enhance the reliability of the warning results; the combination of dynamic weight adjustment and probability model realizes the refined quantification of warning levels, effectively reflecting the probability of disaster occurrence and the degree of impact. The construction of streaming computing framework and structured data sets ensures the real-time processing and efficient transmission of warning information; the priority queue mechanism based on risk scoring ensures that high-risk warning information is pushed first, improving the efficiency of emergency response. The feedback-driven model update mechanism forms a "warning-feedback-optimization" closed loop, enabling the system to dynamically optimize the warning strategy according to the actual disaster situation, and continuously improve the accuracy and adaptability of the warning.

[0128] In one embodiment, Figure 2As shown, the present application also provides a short-term heavy rainfall regional collaborative early warning system based on multi-source fusion data, which may include:

[0129] The data processing module 701 is used to obtain ground rain gauge, radar and satellite data; multi-scale interpolation and normalization are applied to the data to obtain a precipitation monitoring data set containing timestamps, longitude and latitude, and precipitation intensity.

[0130] The precipitation path prediction module 702 is used to preprocess the precipitation monitoring data set and extract precipitation spatial features, and combine the wind field and pressure field data to input the time series prediction model to generate a precipitation path prediction sequence.

[0131] The warning grade generation module 703 is used to process the precipitation path prediction sequence using a graph structure segmentation algorithm, integrate road network and slope characteristics, and generate a high-precision warning area boundary sequence; if the overlap between the warning area boundary sequence and the historical disaster area exceeds a preset threshold, a probability model is used to generate a graded warning sequence.

[0132] The aforementioned regional coordinated early warning system for short-term heavy precipitation, based on multi-source fusion data, consists of three modules: data processing, precipitation path prediction, and warning grade generation. The data processing module collects heterogeneous data from ground rain gauges, radar, and satellite sources. Using multi-scale interpolation algorithms and normalization techniques to address differences in temporal resolution, spatial coverage, and data format, it integrates these data into a structured precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity. The precipitation path prediction module extracts spatial characteristics of precipitation from this dataset through preprocessing. It then combines wind vector direction and pressure gradient data to quantify the influence of wind and pressure fields by calculating the dynamic field coupling coefficient. This fused spatiotemporal characteristics are then input into a time series prediction model to generate a predicted sequence of the movement trajectory of the precipitation core area. The warning grade generation module uses a graph segmentation algorithm to perform topological analysis on the predicted precipitation path sequence. It then integrates geographic information such as road network topological connectivity and terrain slope change rate to generate a high-precision warning area boundary sequence. By calculating the overlap between this sequence and historical disaster areas, and when the overlap exceeds a preset threshold, a probabilistic model is used to analyze disaster parameters, ultimately generating a graded warning sequence. The use of this system can improve the integrity and accuracy of precipitation monitoring data sets, enhance the ability to capture the evolution of precipitation paths, effectively improve the timeliness and reliability of short-term heavy rainfall warnings, and provide accurate data support for disaster prevention and emergency decision-making.

[0133] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0134] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the short-term heavy rainfall regional collaborative early warning method and system based on multi-source fusion data as described above are implemented.

[0135] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0136] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0137] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A short-term heavy rainfall regional collaborative early warning method based on multi-source fusion data, characterized by: The method comprises: Obtaining ground rain gauge, radar, and satellite data; applying multi-scale interpolation and normalization to the data to obtain a precipitation monitoring dataset containing timestamps, longitude and latitude, and precipitation intensity; Preprocessing the precipitation monitoring data set and extracting precipitation spatial features, combining wind field and pressure field data to input a time series prediction model to generate a precipitation path prediction sequence; Using a graph structure segmentation algorithm to process the precipitation path prediction sequence, integrating road network and slope features, and generating a high-precision warning area boundary sequence; If the overlap between the warning area boundary sequence and the historical disaster area exceeds a preset threshold, a probability model is used to generate a graded warning sequence.

2. The method according to claim 1, characterized in that The data is processed by multi-scale interpolation and normalization to obtain a precipitation monitoring data set containing timestamps, longitude and latitude, and precipitation intensity, including: Obtaining the temporal resolution of ground rain gauge data, the spatial coverage of radar data, and the band characteristics of satellite data; the ground rain gauge data includes precipitation records at discrete points; Generating sliding window parameters for multi-scale interpolation according to the time resolution; the sliding window parameters are used to control data alignment in the time dimension; Determining a grid density threshold of an interpolation algorithm based on the spatial coverage range; wherein the grid density threshold is negatively correlated with a radar scanning radius; Processing the data using a multi-scale interpolation algorithm based on the sliding window parameters and the grid density threshold to obtain an interpolated data set; Calculate the water vapor absorption index according to the band characteristics and obtain a weight factor for standardization; Performing a fusion operation on the interpolation dataset according to the weight factor to obtain a three-dimensional matrix including a timestamp, longitude and latitude, and precipitation intensity; The three-dimensional matrix is projected and transformed with a preset coordinate system to obtain a precipitation monitoring data set.

3. The method according to claim 2, characterized in that After obtaining the precipitation monitoring data set, the method further includes: Dividing the abnormal precipitation area according to the grid density threshold and dynamically iterating and updating the interpolation algorithm to obtain dynamically updated data; Performing time series smoothing on the dynamically updated data using the sliding window parameters to obtain a smoothed data set covering the missing period; Adjusting a normalization coefficient of the normalization process according to the water vapor absorption index; the normalization coefficient acts on intensity unit conversion of different data sources; The projected data are superimposed with the data of the smoothed data set to obtain a final precipitation monitoring data set.

4. The method according to claim 1, wherein The method of preprocessing the precipitation monitoring data set and extracting precipitation spatial features, and combining wind field and pressure field data to input a time series prediction model to generate a precipitation path prediction sequence includes: Obtaining an original precipitation intensity distribution in the precipitation monitoring data set; the original precipitation intensity distribution includes spatial coordinates and timestamp information; Mapping the original precipitation intensity distribution into a spatiotemporal interpolation grid to obtain an interpolated precipitation intensity matrix; the spatiotemporal interpolation grid covers the geographic range of the target area; Acquire the vector direction field in the wind field data and the gradient field in the pressure field data; A dynamic field coupling coefficient is calculated based on the vector direction field and the gradient field; the dynamic field coupling coefficient reflects the association weight of the wind pressure field with the spatial characteristics of precipitation; Performing matrix fusion on the spatial gradient features in the precipitation intensity matrix and the dynamic field coupling coefficient to generate a spatiotemporal feature fusion tensor; the spatial gradient features include the intensity change rate of adjacent grid cells; Inputting the spatiotemporal feature fusion tensor into a time series prediction model to output a probability distribution of precipitation paths for multiple future time steps; the probability distribution of precipitation paths includes a sequence of spatial displacement vectors; A precipitation path prediction sequence is generated according to the spatial displacement vector sequence; the precipitation path prediction sequence marks the movement trajectory of the precipitation core area in the target area.

5. The method according to claim 4, characterized in that The spatiotemporal feature fusion tensor is calculated by the following formula: Among them, T represents the spatiotemporal feature fusion tensor, I represents the spatiotemporal precipitation intensity matrix, G represents the spatial gradient feature matrix, Λ represents the dynamic field coupling coefficient, Attention(·) represents the attention mechanism function, and W α represents the learnable weight matrix, represents the scaling factor, σ represents the GELU activation function, and b represents the bias vector.

6. The method according to claim 1, wherein The method of processing the precipitation path prediction sequence using a graph structure segmentation algorithm, integrating road network and slope features, and generating a high-precision warning area boundary sequence includes: Processing the precipitation path prediction sequence using a graph structure segmentation algorithm to generate a path graph structure including node connection relationships; Acquiring topological connectivity data and slope characteristic data of the road network; the slope characteristic data includes a slope change rate of each grid cell; Determine the neighborhood weight of each node in the path graph structure according to the topological connectivity data, and calculate the path similarity matrix in combination with the slope change rate; Processing the path similarity matrix using a spatial smoothing algorithm to generate a grid risk value distribution containing fusion features; Setting a dynamic threshold of the grid risk value according to the path similarity matrix and the slope change rate weighted; Binarize the grid risk value distribution based on the dynamic threshold to obtain the initial warning area boundary; The initial warning area boundary is optimized by iteratively adjusting the neighborhood weight and smoothing parameter to obtain a high-precision warning area boundary sequence containing a time series.

7. The method according to claim 1, characterized in that The method further comprises: Performing spatial grid coding on the warning area according to the spatial coordinate data of the warning area boundary sequence to obtain a gridded boundary sequence; Extracting a disaster history distribution library that intersects with the gridded boundary sequence from a disaster history database; the disaster history distribution library includes disaster type and impact range parameters; Calculating the regional coincidence index between the gridded boundary sequence and the disaster historical distribution database; the regional coincidence index includes a spatial coverage factor and a temporal frequency factor; When the regional coincidence index exceeds a preset threshold, a dynamic weight parameter adjustment algorithm is used to process the disaster type and impact range parameters to obtain an adjusted dynamic weight parameter; Input the dynamic weight parameters into the probability model and output a graded warning sequence including warning levels and confidence values; Inputting the graded warning sequence into a stream computing framework to obtain a structured warning message dataset; the structured warning message dataset includes a warning level and an associated data source identifier; Matching a preset anomaly detection model according to the warning level to generate a set of abnormal events with risk scores; Input the abnormal event set into the priority queue of the distributed messaging system to generate a real-time warning push sequence carrying the terminal identification and push time; the priority queue dynamically adjusts the message order based on the risk score; According to the associated data source identifier, a preset feedback receiving interface is matched, the parameter weights of the anomaly detection model are updated to process the newly added warning message data set, and an updated anomaly event set is obtained and added to the priority queue.

8. A short-term heavy rainfall regional collaborative early warning system based on multi-source fusion data, characterized by: The system comprises: The data processing module is used to obtain ground rain gauge, radar and satellite data; multi-scale interpolation and normalization are applied to the data to obtain a precipitation monitoring data set containing timestamps, longitude and latitude, and precipitation intensity; A precipitation path prediction module is used to preprocess the precipitation monitoring data set and extract precipitation spatial features, and combine the wind field and pressure field data to input the time series prediction model to generate a precipitation path prediction sequence; The warning grade generation module is used to process the precipitation path prediction sequence using a graph structure segmentation algorithm, integrate road network and slope characteristics, and generate a high-precision warning area boundary sequence; if the overlap between the warning area boundary sequence and the historical disaster area exceeds a preset threshold, a probability model is used to generate a graded warning sequence.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

Images