Meteorological disaster risk grading intelligent evaluation and pushing method for calmness area

Abstract

The invention provides a gradation-oriented meteorological disaster risk grading intelligent evaluation and pushing method, and relates to the technical field of meteorological disaster early warning, and the method comprises the steps: obtaining meteorological monitoring data of a plurality of monitoring points in a town, employing a space-time attention mechanism and a dynamic weight mechanism to extract and fuse space-time features and geographic element features, and obtaining a gradation-oriented meteorological disaster risk grading intelligent evaluation and pushing result; generating a meteorological disaster risk level through a bidirectional long-short-term memory network; meanwhile, a vulnerability evaluation model is constructed based on functional area distribution information, and risk levels and vulnerability features are dynamically fused to obtain a comprehensive risk index; and finally, constructing a distributed storage network through a hierarchical cascade block chain, and carrying out encryption and fragmentation pushing on the early warning information. According to the invention, accurate assessment and efficient early warning of the meteorological disaster risk of the calmweight can be realized, and the reliability and timeliness of early warning information are improved.

Description

Technical Field

[0001] The present invention relates to the field of meteorological disaster early warning technology, and in particular to a method for intelligently assessing and pushing meteorological disaster risk levels for towns. Background Art

[0002] With the acceleration of urbanization, towns, as key nodes at the junction of urban and rural areas, have a direct bearing on regional security and economic development in their ability to defend against meteorological disasters. Currently, meteorological disaster risk assessments in towns primarily rely on data from meteorological stations and manual judgment. Various regions have also established early warning information dissemination systems at varying levels. These systems collect meteorological monitoring data, integrate it with geographic information systems, assess disaster risks, and issue early warning information to relevant departments and the public.

[0003] However, the existing meteorological disaster risk assessment methods still have several shortcomings. Traditional assessment methods often analyze meteorological elements separately, ignore the temporal and spatial correlation between elements, and the way of integrating geographical element characteristics is too simple, making it difficult to accurately depict the evolution of disasters in complex geographical environments; the existing risk assessment model does not fully consider the differences in functional zoning of towns, and the depiction of regional vulnerability characteristics is not accurate enough, and it is impossible to achieve multi-level and multi-dimensional comprehensive risk assessment; there are problems such as insufficient information security, inaccurate terminal screening, and low push efficiency in the process of pushing early warning information, making it difficult to ensure the timely and reliable delivery of early warning information.

[0004] In summary, this paper proposes a meteorological disaster risk assessment method that adaptively integrates meteorological elements and geographic features through a spatiotemporal attention mechanism and scene-aware dynamic weighting. It introduces a graph neural network and a temporal adaptive fusion network to accurately model regional vulnerability. Furthermore, it designs an intelligent push mechanism based on blockchain and a two-layer spatial index to achieve accurate assessment and efficient early warning of meteorological disaster risks in townships. This method can effectively improve the accuracy of risk assessment and the reliability of early warning push, providing important decision-making support for township meteorological disaster prevention. Summary of the Invention

[0005] The embodiment of the present invention provides a method for intelligently assessing and pushing meteorological disaster risk levels in towns, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provides a method for intelligent assessment and push of meteorological disaster risk levels for towns, including:

[0008] Obtain meteorological monitoring data from multiple monitoring points within the town and generate a meteorological monitoring data set;

[0009] Based on the meteorological monitoring dataset, the spatiotemporal features are extracted through the spatiotemporal attention mechanism. The spatiotemporal features are adaptively fused with the preset geographic element features based on the dynamic weight mechanism of scene perception to obtain a fused feature vector. The fused feature vector is processed through a bidirectional long short-term memory network to generate the meteorological disaster risk level.

[0010] Obtain the distribution information of each functional area in the town, generate a regional distribution data set, build a town vulnerability assessment model, and calculate the regional vulnerability assessment index;

[0011] Based on multi-level regional vulnerability assessment indicators, a graph neural network and a time-series adaptive fusion network are constructed to dynamically integrate the meteorological disaster risk level with the regional vulnerability characteristics to obtain a comprehensive risk index. Spatial constraint clustering optimization is performed on the town area to generate risk zoning data.

[0012] A distributed storage network is built based on a hierarchical cascade blockchain. The risk partition data is encrypted to generate an early warning information package. The target early warning terminal is screened by calculating the comprehensive matching degree through a double-layer spatial index. The early warning information package is pushed in segments based on priority sorting.

[0013] In an optional embodiment,

[0014] Based on the meteorological monitoring dataset, the spatiotemporal features are extracted through the spatiotemporal attention mechanism. The spatiotemporal features are adaptively fused with the preset geographic element features based on the dynamic weight mechanism of scene perception to obtain a fused feature vector. The fused feature vector is processed through a bidirectional long short-term memory network to generate the meteorological disaster risk level, including:

[0015] Construct a spatiotemporal attention module consisting of a temporal attention submodule and a spatial attention submodule, extract features from the meteorological monitoring dataset to obtain temporal features and spatial features, and perform weighted fusion calculation on the temporal features and spatial features to obtain spatiotemporal fusion features;

[0016] Performing coding analysis on the current meteorological scene to generate a scene vector; calculating a dynamic fusion weight of the spatiotemporal fusion feature and a preset geographic element feature based on the scene vector, performing feature adaptive selection according to the dynamic fusion weight, and obtaining a fusion feature vector;

[0017] The fused feature vector is input into the forward unit and the reverse unit of the bidirectional long short-term memory network, the forward hidden state and the reverse hidden state are determined and spliced, the risk level probability distribution is obtained by calculating the output weight matrix, and the meteorological disaster risk level is generated.

[0018] In an optional embodiment,

[0019] Performing coding analysis on the current meteorological scene to generate a scene vector; calculating the dynamic fusion weight of the spatiotemporal fusion feature and the preset geographic element feature based on the scene vector, performing feature adaptive selection according to the dynamic fusion weight, and obtaining the fusion feature vector includes:

[0020] Constructing a dynamic scene-aware feature fusion network, comprising a scene encoder and an adaptive feature selector. The scene encoder employs a multi-head self-attention mechanism to interactively model the multidimensional meteorological elements of the current meteorological scene, generating a scene feature matrix. The scene feature matrix is then optimized through residual connections and normalization layers to generate a scene vector.

[0021] Based on the scene vector, a feature importance evaluation module, a feature relevance calculation unit, and a scene adaptability calculation unit are constructed; wherein the feature relevance calculation unit calculates the mutual information score between the spatiotemporal fusion feature and the preset geographic element feature, and the scene adaptability calculation unit calculates the contribution score of each feature to the current scene; the mutual information score and the contribution score are weightedly combined to obtain a feature importance score;

[0022] The adaptive feature selector constructs a soft gating mechanism based on the feature importance score, maps the feature importance score to the feature selection probability through the sigmoid function, dynamically adjusts the feature fusion weight according to the feature selection probability, realizes adaptive selection and fusion of features, and generates a fused feature vector.

[0023] In an optional embodiment,

[0024] Based on the multi-level regional vulnerability assessment indicators, a graph neural network and a time series adaptive fusion network are constructed to dynamically integrate the meteorological disaster risk level with the regional vulnerability characteristics to obtain a comprehensive risk index. Spatial constraint clustering optimization is performed on the town area to generate risk zoning data including:

[0025] Acquiring regional vulnerability data at a current moment according to multi-level regional vulnerability assessment indicators, wherein the multi-level regional vulnerability assessment indicators include a population vulnerability indicator, an economic vulnerability indicator, an infrastructure vulnerability indicator, and an emergency response capability indicator;

[0026] A graph neural network model is constructed, using each indicator in the regional vulnerability data as a graph node feature and constructing graph edge connections based on geographic adjacency. The spatial dependency features between graph node features are extracted through a graph convolutional layer, and the spatial dependency features at different scales are adaptively aggregated using an attention mechanism to obtain a regional vulnerability feature vector.

[0027] A time series adaptive fusion network including a gated recurrent unit and a dynamic calibration module is constructed. The meteorological disaster risk level and the regional vulnerability feature vector are input into the gated recurrent unit to capture the time series fusion features. The dynamic calibration module adaptively calibrates and assesses the credibility of the time series fusion features based on historical data distribution to generate a comprehensive risk index with a confidence interval.

[0028] The comprehensive risk index is input into the spatial constraint clustering algorithm to perform initial risk partitioning, spatial adjacency constraint items are introduced to ensure the spatial continuity of the initial risk partitions, and the boundaries of the initial risk partitions are optimized by the simulated annealing optimization algorithm to obtain risk partition data;

[0029] The comprehensive risk index is input into the spatial constraint objective function for iterative optimization to obtain the initial risk partition. The local density value and the nearest high-density distance value are calculated based on the regional adjacency graph to identify the density peak point. The initial risk partition boundary is optimized with the density peak point to obtain the risk partition data.

[0030] In an optional embodiment,

[0031] The comprehensive risk index is input into the spatial constraint objective function for iterative optimization to obtain the initial risk partition. The local density value and the nearest high-density distance value are calculated based on the regional adjacency graph to identify the density peak point. The initial risk partition boundary is optimized with the density peak point. The risk partition data obtained includes:

[0032] Receive a comprehensive risk index, including risk values of multiple risk sample points;

[0033] The sum of the Euclidean distances from multiple risk sample points to the category center is calculated to obtain the intra-class distance. A spatial adjacency relationship is established based on the geographical locations of the multiple risk sample points, and the category difference values between the adjacent points are calculated to obtain a spatial adjacency constraint. The intra-class distance and the spatial adjacency constraint are combined to construct a spatial constraint objective function.

[0034] The comprehensive risk index is input into the spatial constraint objective function, and the category center point is obtained through iterative optimization. Based on the category center point, spatial constraint clustering calculation is performed on multiple risk sample points to obtain the initial risk partition result;

[0035] Constructing a regional adjacency graph based on the initial risk partitioning result, calculating the number of adjacent points within a preset cutoff distance for each risk sample point to obtain a first local density value set, selecting a target risk sample point from the first local density value set, determining adjacent risk sample points that are adjacent to the target risk sample point and have a local density value greater than the target risk sample point, calculating the minimum distance value from the target risk sample point to the adjacent risk sample point, and obtaining a nearest high-density distance value;

[0036] A density peak point in the initial risk partition result is identified based on the local density value and the nearest high density distance value, and a boundary optimization is performed on the initial risk partition result based on the density peak point to obtain risk partition data.

[0037] In an optional embodiment,

[0038] Based on the hierarchical cascade blockchain, a distributed storage network is built to encrypt risk partition data to generate early warning information packages. The target early warning terminals are screened through the comprehensive matching degree calculated by the double-layer spatial index. The early warning information packages are pushed in fragments based on priority sorting, including:

[0039] Constructing a hierarchical and cascaded blockchain network, establishing a distributed storage structure, and generating early warning information blocks in the distributed storage structure;

[0040] Generate warning information based on risk partition data, asymmetrically encrypt the warning information to obtain warning information encrypted data, combine it with digital signature information to generate a warning information package, and store it in the warning information block;

[0041] Constructing a two-layer spatial index tree, calculating the comprehensive matching degree between the warning range in the warning information package and the warning terminal based on the multi-dimensional matching index, and screening the target warning terminal according to the comprehensive matching degree;

[0042] Mapping the risk level in the warning information package to a push priority to obtain priority data, and prioritizing the target warning terminals based on the priority data to obtain a warning terminal push sequence;

[0043] According to the early warning terminal push sequence, the early warning information package is fragmented and stored to obtain early warning information fragment data, and the early warning information package is pushed to the target early warning terminal through dynamic load balancing.

[0044] In an optional embodiment,

[0045] Construct a two-layer spatial index tree, calculate the comprehensive matching degree between the warning range in the warning information package and the warning terminal based on the multi-dimensional matching index, and select the target warning terminals according to the comprehensive matching degree, including:

[0046] In the two-layer spatial index tree, the first layer constructs regional index nodes based on administrative regions, and the second layer constructs location index nodes based on geographic coordinates;

[0047] Divide the area into grid cells according to the warning range, and retrieve candidate warning terminals within the grid cells using a two-layer spatial index tree; construct a boundary polygon of the warning range, and calculate the distance from the candidate warning terminal to the boundary polygon to obtain the spatial distance matching degree;

[0048] Perform spatial calculation on the warning range and the administrative area boundary to obtain the overlapping area, and calculate the regional matching degree based on the overlapping area; extract the warning information reception delay and data integrity rate based on the historical warning data of the candidate warning terminal to obtain the terminal reliability index;

[0049] Constructing a traffic network topology structure, calculating the shortest path from the candidate warning terminal to the warning range to obtain path connectivity; obtaining population distribution data within the warning range, calculating the number of people covered by the candidate warning terminal to obtain a coverage index;

[0050] The spatial distance matching degree, the regional matching degree, the terminal reliability index, the path connectivity and the coverage index are normalized to determine a comprehensive score, and the warning terminal with the highest comprehensive score is selected as the target warning terminal.

[0051] In an embodiment of the present invention, by acquiring meteorological monitoring data from multiple monitoring points and generating a data set, real-time monitoring and assessment of meteorological disasters in a town can be achieved, thereby improving the response speed and accuracy to meteorological changes; the spatiotemporal attention mechanism and the bidirectional long short-term memory network are used to process meteorological characteristics, which can adaptively integrate the characteristics of geographical elements to generate a more accurate meteorological disaster risk level and enhance the scientific nature of risk assessment; the distributed storage network constructed based on the hierarchical cascade blockchain can safely and efficiently generate and push early warning information packages, ensure the timeliness and reliability of early warning information, and enhance the ability of the town to respond to meteorological disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a method for intelligently assessing and pushing meteorological disaster risk levels for towns according to an embodiment of the present invention;

[0053] Figure 2 This is a diagram showing the effect of feature importance evaluation under different meteorological scenarios;

[0054] Figure 3 A comparison chart of technical algorithm performance and resource consumption selected for the early warning terminal. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0056] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0057] Figure 1 FIG. 1 is a flow chart of a method for intelligently assessing and pushing meteorological disaster risk levels in towns according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] Obtain meteorological monitoring data from multiple monitoring points within the town and generate a meteorological monitoring data set;

[0059] Based on the meteorological monitoring dataset, the spatiotemporal features are extracted through the spatiotemporal attention mechanism. The spatiotemporal features are adaptively fused with the preset geographic element features based on the dynamic weight mechanism of scene perception to obtain a fused feature vector. The fused feature vector is processed through a bidirectional long short-term memory network to generate the meteorological disaster risk level.

[0060] Obtain the distribution information of each functional area in the town, generate a regional distribution data set, build a town vulnerability assessment model, and calculate the regional vulnerability assessment index;

[0061] Based on multi-level regional vulnerability assessment indicators, a graph neural network and a time-series adaptive fusion network are constructed to dynamically integrate the meteorological disaster risk level with the regional vulnerability characteristics to obtain a comprehensive risk index. Spatial constraint clustering optimization is performed on the town area to generate risk zoning data.

[0062] A distributed storage network is built based on a hierarchical cascade blockchain. The risk partition data is encrypted to generate an early warning information package. The target early warning terminal is screened by calculating the comprehensive matching degree through a double-layer spatial index. The early warning information package is pushed in segments based on priority sorting.

[0063] In an optional embodiment, based on a meteorological monitoring dataset, spatiotemporal features are extracted through a spatiotemporal attention mechanism, and the spatiotemporal features are adaptively fused with preset geographic element features based on a scene-aware dynamic weighting mechanism to obtain a fused feature vector. The fused feature vector is processed through a bidirectional long short-term memory network to generate a meteorological disaster risk level including:

[0064] Construct a spatiotemporal attention module consisting of a temporal attention submodule and a spatial attention submodule, extract features from the meteorological monitoring dataset to obtain temporal features and spatial features, and perform weighted fusion calculation on the temporal features and spatial features to obtain spatiotemporal fusion features;

[0065] Performing coding analysis on the current meteorological scene to generate a scene vector; calculating a dynamic fusion weight of the spatiotemporal fusion feature and a preset geographic element feature based on the scene vector, performing feature adaptive selection according to the dynamic fusion weight, and obtaining a fusion feature vector;

[0066] The fused feature vector is input into the forward unit and the reverse unit of the bidirectional long short-term memory network, the forward hidden state and the reverse hidden state are determined and spliced, the risk level probability distribution is obtained by calculating the output weight matrix, and the meteorological disaster risk level is generated.

[0067] A spatiotemporal attention module is constructed, consisting of a temporal attention submodule and a spatial attention submodule. The temporal attention submodule processes the time series data in the meteorological monitoring dataset and extracts features along the temporal dimension. Specifically, the meteorological monitoring dataset is arranged chronologically to form a time series matrix, where each row represents meteorological data at a specific time point and each column represents a meteorological indicator. The temporal attention submodule calculates the correlations between different time points to generate a temporal attention weight matrix, which reflects the importance of each time point to the current prediction task.

[0068] The spatial attention submodule processes the spatially distributed data in the meteorological monitoring dataset and extracts spatial features. The meteorological monitoring data is gridded according to geographic location, forming a spatial grid matrix, where each grid cell contains the meteorological data for that location. The spatial attention submodule calculates the spatial correlation between different grid cells and generates a spatial attention weight matrix, which reflects the importance of different geographic locations to the current prediction task.

[0069] The temporal and spatial attention weight matrices are multiplied by the raw meteorological monitoring data to obtain weighted temporal and spatial features, respectively. Then, the temporal and spatial features are fused together through a weighted fusion calculation to obtain spatiotemporal fusion features. During the fusion process, the weight ratio of temporal and spatial features can be adjusted according to actual needs, for example, a temporal weight of 0.6 and a spatial weight of 0.4.

[0070] The current meteorological scene is encoded and analyzed to generate a scene vector. The scene vector contains key information about the current meteorological scene, such as weather type, season, and geographic location. Specifically, various attributes of the current meteorological scene are encoded. For example, the weather type (sunny, rainy, snowy, etc.) is encoded as a one-hot vector, the season (spring, summer, autumn, winter) is encoded as a one-hot vector, and the geographic location (longitude, latitude) is normalized. These encoded attributes are then concatenated to form the scene vector.

[0071] Based on the scene vector, dynamic fusion weights are calculated for spatiotemporal fusion features and pre-set geographic element features. Geographic element features include terrain, altitude, vegetation cover, and other geographic information. The dynamic fusion weight calculation process is as follows: First, the scene vector and geographic element features are concatenated to form an input vector. Then, the input vector is processed through a multi-layer perceptron network to generate dynamic fusion weights. These weights reflect the importance of spatiotemporal fusion features and geographic element features to the prediction results under different meteorological scenarios.

[0072] Adaptive feature selection is performed based on the dynamic fusion weights to produce a fused feature vector. Specifically, the spatiotemporal fusion features and geographic feature features are weighted and summed according to the dynamic fusion weights to produce the fused feature vector. For example, in a rainstorm, the dynamic fusion weights may favor geographic feature features, while in clear weather, the dynamic fusion weights may favor spatiotemporal fusion features.

[0073] The fused feature vector is input into a bidirectional long short-term memory network for processing. The bidirectional long short-term memory network comprises forward units and reverse units. The forward units process the fused feature vector in chronological order, while the reverse units process the fused feature vector in reverse chronological order. The forward units generate forward hidden states, while the reverse units generate reverse hidden states.

[0074] The forward hidden state and the reverse hidden state are concatenated to form a complete hidden state representation. The hidden state is then mapped to the risk level probability distribution space using the output weight matrix. The dimension of the output weight matrix is the hidden state dimension multiplied by the number of risk levels. The product of the output weight matrix and the hidden state is converted into a probability distribution using the softmax function to obtain the probability of each risk level.

[0075] Based on the probability distribution, the risk level with the highest probability is selected as the final prediction result. For example, the risk level can be divided into four levels: low risk, medium risk, high risk, and very high risk. The corresponding probabilities are 0.1, 0.3, 0.4, and 0.2, respectively. The final prediction result is high risk.

[0076] For example, for a meteorological disaster risk forecast for a specific region, the meteorological monitoring dataset contains meteorological data for the past 24 hours, with a sampling interval of one hour. Each time point contains 10 meteorological indicators, including temperature, humidity, wind speed, and precipitation. The geographic location is divided into 100 grid cells of 10 km x 10 km.

[0077] After processing by the temporal attention submodule, a 24×24 temporal attention weight matrix is generated, where the weight value in row 12, column 8 is 0.15, indicating that the data from 8 hours ago contributes 15% to the current prediction. After processing by the spatial attention submodule, a 100×100 spatial attention weight matrix is generated, where the weight value in row 45, column 67 is 0.08, indicating that the data in the 67th grid cell contributes 8% to the prediction of the 45th grid cell.

[0078] The scenario vector contains information such as the current weather type (rainy), season (summer), and geographic location (longitude 120.5, latitude 30.2). After dynamic fusion weight calculation, the weight of the spatiotemporal fusion feature is 0.7, and the weight of the geographic element feature is 0.3, indicating that in the current scenario, the spatiotemporal fusion feature has a greater impact on the prediction results.

[0079] After processing by the bidirectional long short-term memory network, the generated risk level probability distribution is: low risk 0.05, medium risk 0.25, high risk 0.45, and extremely high risk 0.25. The final prediction result is high risk.

[0080] In this embodiment, through independent temporal attention submodules and spatial attention submodules, the temporal dynamic change patterns and spatial distribution characteristics of meteorological data are extracted respectively, avoiding key information that may be missed in simple splicing or convolution; a scene vector is generated based on the current meteorological scene encoding to adjust the fusion weights between spatiotemporal features and preset geographical elements, so that the model can adaptively select the most discriminative features for different climatic conditions and terrain environments; dynamic fusion weights and adaptive selection mechanisms can highlight key features and suppress irrelevant or noise information, thereby improving the model's sensitivity and robustness to extreme weather or emergencies; through forward and reverse hidden state splicing and output weight matrices, the model can provide a more refined and reliable probability distribution of disaster risk levels, improve the accuracy of early warnings, and assist decision makers in targeted disaster prevention and mitigation deployments.

[0081] In an optional embodiment, encoding analysis is performed on the current meteorological scene to generate a scene vector; dynamic fusion weights of the spatiotemporal fusion features and preset geographic element features are calculated based on the scene vector, and feature adaptive selection is performed according to the dynamic fusion weights to obtain a fusion feature vector, including:

[0082] Constructing a dynamic scene-aware feature fusion network, comprising a scene encoder and an adaptive feature selector. The scene encoder employs a multi-head self-attention mechanism to interactively model the multidimensional meteorological elements of the current meteorological scene, generating a scene feature matrix. The scene feature matrix is then optimized through residual connections and normalization layers to generate a scene vector.

[0083] Based on the scene vector, a feature importance evaluation module, a feature relevance calculation unit, and a scene adaptability calculation unit are constructed; wherein the feature relevance calculation unit calculates the mutual information score between the spatiotemporal fusion feature and the preset geographic element feature, and the scene adaptability calculation unit calculates the contribution score of each feature to the current scene; the mutual information score and the contribution score are weightedly combined to obtain a feature importance score;

[0084] The adaptive feature selector constructs a soft gating mechanism based on the feature importance score, maps the feature importance score to the feature selection probability through the sigmoid function, dynamically adjusts the feature fusion weight according to the feature selection probability, realizes adaptive selection and fusion of features, and generates a fused feature vector.

[0085] In a specific embodiment, the scene encoder uses a multi-head self-attention mechanism to interactively model the multi-dimensional meteorological elements of the current meteorological scene. The multi-head self-attention mechanism projects the input meteorological element representation into three spaces: query, key, and value, and each head is responsible for capturing the correlation between different dimensions. In the specific implementation, it is assumed that the input meteorological element set is X, with dimensions (N, D), where N is the number of samples and D is the feature dimension. X is projected into query Q, key K, and value V through linear projection matrices Wq, Wk, and Wv, with dimensions (N, H, D / H), respectively, where H is the number of attention heads. In practical applications, H can be set to 8 to balance computational efficiency and expression ability.

[0086] For each attention head h, an attention weight is calculated by performing matrix multiplication on the query Q and the transpose of the key K, dividing it by a scaling factor (usually the square root of the key vector dimension), and then applying a softmax function to obtain the attention weight. For example, when the input consists of five meteorological elements such as temperature, humidity, and air pressure from 10 weather stations, the original input dimension is (10, 5). After multi-head self-attention calculation, the interrelationships between different meteorological elements can be captured.

[0087] Multiply the attention weights by the value V to obtain the output of each head, concatenate the outputs of all heads, and then perform linear projection to obtain the final scene feature matrix. For example, when using 8 attention heads, the output dimension of each head is (10, 5 / 8), and after concatenation it is (10, 5). Projection can maintain the original dimension unchanged.

[0088] The scene feature matrix is optimized through residual connections and normalization layers. Residual connections add the original input to the self-attention output, effectively alleviating the vanishing gradient problem in deep networks. Layer normalization then standardizes the results to stabilize the training process. In practice, layer normalization calculates the mean and variance of each sample feature, normalizes the features, and then applies learnable scaling and offset parameters. For example, for a feature matrix of dimension (10, 5), the mean and variance are calculated along the feature dimension to ensure a stable data distribution. After these processes, a scene vector of dimension D is generated, which serves as a compact representation of the scene.

[0089] Based on the scene vector, a feature importance evaluation module is constructed, including a feature correlation calculation unit and a scene adaptability calculation unit. The feature correlation calculation unit calculates the mutual information score between the spatiotemporal fusion feature and the preset geographic element feature. Mutual information measures the degree of mutual dependence between two variables, and the higher the value, the stronger the correlation. A KNN (K nearest neighbor)-based method is used to estimate mutual information. First, the feature space is divided into a grid, the histograms of the joint distribution and the marginal distribution are calculated, and then the mutual information is calculated. For example, for 10 geographic elements such as rivers and mountains, the mutual information between meteorological features and each geographic element is calculated to obtain a mutual information score vector with a dimension of 10.

[0090] The scene adaptability calculation unit calculates the contribution score of each feature to the current scene. Using an attention mechanism, the scene vector is used as a query and attention weights are calculated for each feature vector. A softmax function based on a temperature parameter is designed to adjust the smoothness of the weight distribution. The temperature parameter is typically set between 0.5 and 2, with lower temperatures resulting in a more concentrated distribution and higher temperatures resulting in a more uniform distribution. For example, when predicting a heavy rainfall scenario, the pressure gradient feature may receive a high contribution score of 0.72, while in a clear weather scenario, the feature score may be only 0.15.

[0091] The mutual information score and contribution score are weighted together to produce a feature importance score. The weighting coefficients can be optimized using the validation set; typical settings are a mutual information score weight of 0.4 and a contribution score weight of 0.6. Feature importance scores typically range from 0 to 1, with higher values indicating greater feature importance. For example, in a typhoon scenario, wind speed and air pressure features might receive a high importance score of 0.85 or above.

[0092] The adaptive feature selector implements a soft gating mechanism based on feature importance scores. Feature importance scores are mapped to feature selection probabilities using a sigmoid function. The output of the sigmoid function ranges from 0 to 1, representing the probability of selecting the corresponding feature. To enhance differentiation, a temperature parameter can be introduced to adjust the steepness of the sigmoid function, with a typical value of 0.5. For example, a feature with an importance score of 0.8 has a selection probability of approximately 0.69 after the sigmoid transformation.

[0093] Dynamically adjust feature fusion weights based on feature selection probability. Use selection probability directly as a weight or further map it through a gating function. In practical applications, a minimum weight threshold (e.g., 0.05) can be set to ensure that all features have a chance to participate in fusion, avoiding complete information loss. Features with high selection probabilities, such as the pressure gradient feature with a probability of 0.85, are given a larger weight; features with low probabilities, such as the sunshine duration feature with a probability of 0.2, are given a smaller weight.

[0094] The original features are multiplied by the corresponding weights and summed to achieve adaptive feature selection and fusion, generating a fused feature vector. For example, for 10 feature vectors of dimension 256, after weighted fusion, a fused feature vector of dimension 256 is obtained, which serves as the input for subsequent tasks.

[0095] Existing technologies usually use fixed weights or simple attention mechanisms in meteorological feature fusion, and are unable to dynamically adjust feature selection strategies according to different meteorological scenarios. For example, traditional methods use the same feature fusion strategy under different weather systems, resulting in the weakening of key information or interference from irrelevant information. The starting point of the improvement of this application is to build a system that can perceive the current meteorological scene and dynamically select features based on it, extract scene features through a scene encoder, and design a dual evaluation mechanism of mutual information and scene adaptability to accurately quantify the importance of each feature. A soft gating mechanism is introduced to achieve smooth transition feature selection, avoiding the instability that may be caused by hard gating. Experimental results show that in complex meteorological scenarios such as severe convection and typhoons, this method has an accuracy improvement of 12.3% compared to the fixed weight fusion method and 7.6% compared to the simple attention mechanism, especially showing stronger stability and adaptability in extreme weather forecasting.

[0096] like Figure 2As shown, the feature importance score distribution of the feature importance evaluation module in the scheme of this embodiment in different meteorological scenarios is shown. The chart presents the dynamic changes in the importance of 9 key meteorological and geographical elements in 6 typical meteorological scenarios in the form of a line graph. It can be clearly seen from the data that the importance of each element changes significantly with the scene, reflecting the scene adaptation characteristics of this scheme. In the sunny scene, the temperature feature shows the highest importance (0.872), and the urban heat island effect (0.732) also has a high influence; in the rainfall scene, the precipitation feature reaches the highest importance (0.923), followed by humidity (0.891); in the typhoon scene, the wind speed (0.943) and wind direction (0.923) features are significantly dominant, and air pressure (0.912) is also a key factor; in the haze scene, the urban heat island effect (0.853) and cloud cover (0.812) show a stronger correlation. This dynamically adjusted feature importance evaluation mechanism fully demonstrates that the adaptive feature selector in this embodiment can accurately identify key factors according to the characteristics of different scenarios and optimize the feature fusion weights accordingly, thereby greatly improving the prediction accuracy and model adaptability, especially under extreme or complex meteorological conditions.

[0097] In an optional embodiment, a graph neural network and a time series adaptive fusion network are constructed based on multi-level regional vulnerability assessment indicators. The meteorological disaster risk level is dynamically integrated with the regional vulnerability characteristics to obtain a comprehensive risk index. Spatial constraint clustering optimization is performed on the town area to generate risk zoning data including:

[0098] Acquiring regional vulnerability data at a current moment according to multi-level regional vulnerability assessment indicators, wherein the multi-level regional vulnerability assessment indicators include a population vulnerability indicator, an economic vulnerability indicator, an infrastructure vulnerability indicator, and an emergency response capability indicator;

[0099] A graph neural network model is constructed, using each indicator in the regional vulnerability data as a graph node feature and constructing graph edge connections based on geographic adjacency. The spatial dependency features between graph node features are extracted through a graph convolutional layer, and the spatial dependency features at different scales are adaptively aggregated using an attention mechanism to obtain a regional vulnerability feature vector.

[0100] A time series adaptive fusion network including a gated recurrent unit and a dynamic calibration module is constructed. The meteorological disaster risk level and the regional vulnerability feature vector are input into the gated recurrent unit to capture the time series fusion features. The dynamic calibration module adaptively calibrates and assesses the credibility of the time series fusion features based on historical data distribution to generate a comprehensive risk index with a confidence interval.

[0101] The comprehensive risk index is input into the spatial constraint clustering algorithm to perform initial risk partitioning, spatial adjacency constraint items are introduced to ensure the spatial continuity of the initial risk partitions, and the boundaries of the initial risk partitions are optimized by the simulated annealing optimization algorithm to obtain risk partition data;

[0102] The comprehensive risk index is input into the spatial constraint objective function for iterative optimization to obtain the initial risk partition. The local density value and the nearest high-density distance value are calculated based on the regional adjacency graph to identify the density peak point. The initial risk partition boundary is optimized with the density peak point to obtain the risk partition data.

[0103] Current regional vulnerability data is obtained using multi-level regional vulnerability assessment indicators. These indicators include: demographic vulnerability indicators, which include factors such as population density, proportion of elderly people, and proportion of children. For example, District A in a city has a population density of 5,200 people per square kilometer, an elderly population proportion of 18.3%, and a child population proportion of 15.7%. Economic vulnerability indicators include factors such as per capita GDP, industrial structure, and fiscal revenue. For example, District B in a city has a per capita GDP of 82,000 yuan, a tertiary industry proportion of 65.3%, and annual fiscal revenue of 2.47 billion yuan. Infrastructure vulnerability indicators include factors such as road density, power grid coverage, and communication network coverage. For example, District C in a city has a road density of 4.8 kilometers per square kilometer, a power grid coverage rate of 98.7%, and a communication network coverage rate of 99.3%. Emergency response capacity indicators include factors such as the density of emergency shelters, accessibility to medical resources, and coverage of fire stations. For example, the density of emergency shelters in District D of a certain city is 0.08 per square kilometer, the accessibility of medical resources is 92.5%, and the 10-minute response coverage rate of fire stations is 94.2%.

[0104] A graph neural network model is constructed to process regional vulnerability data. Each indicator in the regional vulnerability data is used as a graph node feature, and graph edge connections are constructed based on geographic adjacency. For adjacent administrative regions, edges are set in the graph; for non-adjacent regions, no edges are set. For example, if Districts A and B in a city are geographically adjacent, an edge AB is set in the graph.

[0105] Graph neural networks use a two-layer graph convolutional architecture to extract spatially dependent features. The first layer uses 32 convolution kernels, and the second layer uses 64 convolution kernels. The activation function is the ReLU function. The graph convolution operation updates the feature representation of the central node by aggregating feature information from neighboring nodes, thereby capturing the spatial dependencies between nodes. For example, a population density feature of 5200 representing a region might be updated to a feature value of 5380 reflecting the influence of the surrounding area after being aggregated with features from neighboring nodes in the graph convolution layer.

[0106] To handle spatially dependent features at different scales, a multi-head attention mechanism is introduced. Eight attention heads are set, each responsible for learning different aspects of node relationships. The features of different nodes are weighted and aggregated using attention coefficients to produce a more representative regional vulnerability feature vector. For example, for District C in City, its infrastructure vulnerability feature may receive a higher attention weight of 0.32 than other features under the multi-head attention mechanism.

[0107] A time-series adaptive fusion network was constructed, consisting of a gated recurrent unit (GRU) and a dynamic calibration module. The meteorological disaster risk level (e.g., extreme precipitation risk level 4) was concatenated with the regional vulnerability feature vector (e.g., the feature vector for city zone A [0.82, 0.65, 0.93, 0.76]) and then fed into the GRU. The GRU contains 128 hidden units and captures time-series fusion features through an update gate and reset gate mechanism.

[0108] The dynamic calibration module adaptively calibrates time series fusion features based on historical data distribution. It constructs a feature distribution library containing 100 historical samples and calculates the deviation between the current fused features and the historical distribution. For example, if the current feature value is 0.85, the historical average for the same period is 0.78, and the standard deviation is 0.05, the normalized deviation is 1.4. Adaptive calibration is performed based on the normalized deviation, and 1000 iterations of Monte Carlo sampling are performed to generate a comprehensive risk index with a 95% confidence interval. For example, the comprehensive risk index for a region is 0.83, and the 95% confidence interval is [0.79, 0.87].

[0109] The comprehensive risk index was input into a spatially constrained clustering algorithm for initial risk zoning. The K-means method was used to categorize the regions into five initial risk levels. Spatial adjacency constraints were then introduced to ensure spatial continuity of the risk zoning. These constraints were defined using an adjacency matrix, with a constraint weight of 0.3 for adjacent regions and 0 for non-adjacent regions. An iterative optimization process (maximum 200 iterations, convergence threshold 0.001) was used to ensure spatial clustering of the initial risk zoning.

[0110] To further optimize the risk partition boundaries, a simulated annealing algorithm was used with an initial temperature of 100 and a cooling coefficient of 0.95. The energy function value was calculated for each boundary region. For example, if the current energy value of a boundary region was 45.8 and the adjusted energy value was 42.3, the adjustment was accepted according to the annealing criterion. After 500 iterations, the risk partition data with optimized boundaries was obtained.

[0111] Optionally, the risk zoning optimization method is to optimize the partition boundaries based on the density peak, construct a regional adjacency graph, and calculate the local density value and the nearest high-density distance value of each region. The local density is calculated by counting the number of adjacent regions whose risk index difference with the target area is less than a threshold of 0.1; the nearest high-density distance is calculated by finding the area with a higher local density than the target area and the closest risk index. For example, the local density value of a region is 4, and the nearest high-density distance value is 0.25. The product exceeds the threshold of 1.0 and is identified as a density peak point. With these density peak points as the center, the initial risk partition boundaries are optimized through the regional attribution allocation algorithm to obtain the final risk partition data.

[0112] In this embodiment, four indicators, namely population, economy, infrastructure and emergency response capacity, are introduced at the same time, which can comprehensively characterize regional vulnerability from multiple dimensions such as social, economic, physical and response capabilities, avoiding the one-sidedness of a single indicator; through the graph convolution layer constructed based on geographic adjacency, the spatial correlation between adjacent regions is effectively captured, so that the vulnerability characteristics not only reflect the situation of individual nodes themselves, but also integrate the influence of the surrounding environment, thereby improving the precision of feature expression; the attention mechanism is used to perform weighted aggregation of spatial dependency features at different scales, automatically adjust the importance according to regional characteristics, highlight key areas, and suppress noise or redundant features. The model can enhance the adaptability of the model to features at different spatial resolutions by combining residual information; the meteorological disaster risk level and vulnerability characteristics are input into the gated recurrent unit (GRU) to capture the temporal evolution process; the dynamic calibration module corrects the fused features based on the historical distribution and gives a confidence interval, which not only enhances the accuracy of the prediction but also provides an uncertainty measure, facilitating the balance between risk and confidence when making decisions; the initial partitioning adopts spatial constraint clustering to ensure that adjacent areas are classified into the same risk level, thereby improving the continuity and operability of the partitioning; the partition boundaries are subsequently optimized through simulated annealing to further smooth the boundaries and improve the rationality of the partitioning.

[0113] In an optional embodiment, the comprehensive risk index is input into the spatial constraint objective function for iterative optimization to obtain an initial risk partition, local density values and nearest high-density distance values are calculated based on the regional adjacency graph to identify density peak points, and the initial risk partition boundaries are optimized using the density peak points. The obtained risk partition data includes:

[0114] Receive a comprehensive risk index, including risk values of multiple risk sample points;

[0115] The sum of the Euclidean distances from multiple risk sample points to the category center is calculated to obtain the intra-class distance. A spatial adjacency relationship is established based on the geographical locations of the multiple risk sample points, and the category difference values between the adjacent points are calculated to obtain a spatial adjacency constraint. The intra-class distance and the spatial adjacency constraint are combined to construct a spatial constraint objective function.

[0116] The comprehensive risk index is input into the spatial constraint objective function, and the category center point is obtained through iterative optimization. Based on the category center point, spatial constraint clustering calculation is performed on multiple risk sample points to obtain the initial risk partition result;

[0117] Constructing a regional adjacency graph based on the initial risk partitioning result, calculating the number of adjacent points within a preset cutoff distance for each risk sample point to obtain a first local density value set, selecting a target risk sample point from the first local density value set, determining adjacent risk sample points that are adjacent to the target risk sample point and have a local density value greater than the target risk sample point, calculating the minimum distance value from the target risk sample point to the adjacent risk sample point, and obtaining a nearest high-density distance value;

[0118] A density peak point in the initial risk partition result is identified based on the local density value and the nearest high density distance value, and a boundary optimization is performed on the initial risk partition result based on the density peak point to obtain risk partition data.

[0119] In one embodiment, comprehensive risk index data is received, which includes risk values for multiple risk sample points. For example, in a city flood risk assessment, risk values for 1,000 sampling points are collected. Each sampling point includes latitude and longitude coordinates and a corresponding flood risk index value. The risk index value ranges from 0 to 100, with larger values indicating higher risk.

[0120] Construct a spatial constraint objective function consisting of two parts: intra-class distance and spatial adjacency constraints. The intra-class distance is calculated by calculating the Euclidean distance from each risk point to the center of its class and summing the distances across all points. For example, assuming there are five classes, each with a center, calculate the distance from each point to the center of its class and sum all distances to obtain the total intra-class distance.

[0121] Spatial adjacency constraints are constructed based on the geographic location of risk sample points. To establish spatial adjacency based on geographic location, the K-nearest neighbor method can be used to determine neighboring points. For example, for each sample point, the eight closest points are selected as its neighbors. The category difference between the adjacent points is then calculated. If two adjacent points belong to different categories, the difference is 1; otherwise, it is 0. The spatial adjacency constraint value is calculated by summing the difference values of all pairs of adjacent points.

[0122] The spatial constraint objective function is constructed by weighting the intra-class distance and spatial adjacency constraints. For example, if the weight coefficients α = 0.7 and β = 0.3, the objective function value = α × intra-class distance + β × spatial adjacency constraint value. The goal is to minimize this function value so that the sample points within the class are clustered and spatially continuous.

[0123] After the comprehensive risk index is input into the spatial constraint objective function, the optimal category center point is solved through an iterative optimization algorithm. Specifically, the improved K-means algorithm can be used: first, K category centers are randomly initialized (for example, K = 5), and then the following steps are repeated until convergence: (1) each sample point is assigned to the nearest category center point; (2) considering the spatial adjacency constraint, the category assignment of the sample point is adjusted; (3) the coordinates of each category center point are recalculated. The algorithm converges when the change in the objective function value between two consecutive iterations is less than a preset threshold (for example, 0.001).

[0124] After iterative optimization is complete, spatially constrained clustering is performed on multiple risk sample points based on the final category center point to obtain the initial risk zoning results. For example, in the risk zoning of a certain city, five risk level areas may be obtained: very low risk area, low risk area, medium risk area, high risk area, and very high risk area.

[0125] Based on the initial risk zoning results, a regional adjacency graph is constructed. For each risk sample point, all its neighboring points within a preset cutoff distance are determined. For example, if the cutoff distance is set to 500 meters, the number of neighboring points within 500 meters of each sample point is counted to obtain the first set of local density values.

[0126] Select target risk sample points from the first local density value set for processing. For each target risk sample point, determine all adjacent risk sample points that are adjacent to it and have a local density value greater than the target point. For example, the local density value of a target point is 15 (indicating that there are 15 adjacent points within 500 meters), and find all points whose local density values are greater than 15 among its adjacent points. Then, calculate the minimum distance value from the target point to these high-density adjacent points, which is the nearest high-density distance value. If there is no adjacent point with a higher local density value, then set the distance value to the maximum distance value in the area.

[0127] Density peaks are identified based on local density values and the distance to the nearest high density. The specific method is to calculate the density decision value for each sample point = local density value × distance to the nearest high density value, and select the points with the highest density decision values as density peaks. For example, within a certain city risk zone, seven density peaks may be identified, representing the centers of areas with different risk characteristics.

[0128] The boundaries of the initial risk partitioning results are optimized based on the density peak points. Each density peak point is used as a seed point to construct a new risk partition. For non-peak sample points, the comprehensive distance to each density peak point is calculated (taking into account the geographical distance and the similarity of risk characteristics), and they are assigned to the partition where the density peak point with the smallest comprehensive distance is located. The sample points at the partition boundary are smoothed to eliminate isolated points and fragmented areas.

[0129] When optimizing flood risk zoning for a city, the initial zoning may have problems such as uneven boundaries and excessive small fragmented areas. Optimizing density peaks results in more continuous and smooth risk zones with boundaries that better reflect actual geographic features, such as those along natural boundaries like rivers and mountains.

[0130] The final output is risk zoning data, including the risk zoning category to which each risk sample point belongs, the spatial extent of each risk zone, and the risk level. For example, the output data may include: sample point ID, longitude and latitude coordinates, risk index value, risk zone category, and other information to facilitate subsequent risk management and disaster prevention and mitigation decision support.

[0131] Through the above method, the risk values and spatial distribution characteristics of risk sample points can be effectively combined to achieve accurate division of risk areas, providing a scientific basis for risk management and disaster prevention and mitigation.

[0132] For example, there are the following datasets: City A dataset (sample size: 1,245; risk areas: 8; spatial range: 450km 2 ; risk index range: 0.12-0.95), City B dataset (sample size: 2,873; risk areas: 12; spatial range: 820 km 2 ; risk index range: 0.08-0.89), City C dataset (sample size: 1,756; risk areas: 9; spatial range: 625 km 2 ; risk index range: 0.15-0.92), comprehensive test set (sample size: 5,874; risk areas: 29; spatial range: 1,895 km 2 ; Risk index range: 0.08-0.95);

[0133]

[0134] As shown in the table above, this comprehensive risk assessment experimental data comparison table comprehensively demonstrates the performance comparison results of this technical solution and eight mainstream clustering algorithms on four different data sets. Four test data set features were used: City A data set (sample size: 1,245; risk areas: 8; spatial range: 450km) 2 ; risk index range: 0.12-0.95), City B dataset (sample size: 2,873; risk areas: 12; spatial range: 820 km 2 ; risk index range: 0.08-0.89), City C dataset (sample size: 1,756; risk areas: 9; spatial range: 625 km 2; risk index range: 0.15-0.92), comprehensive test set (sample size: 5,874; risk areas: 29; spatial range: 1,895 km 2 The data show the performance of each algorithm on four key metrics (accuracy, silhouette coefficient, spatial continuity, and computation time). The data shows that this technical solution leads in all three core metrics: accuracy, silhouette coefficient, and spatial continuity. Specifically, on the comprehensive test set, it achieved 91.5% accuracy, a silhouette coefficient of 0.81, and a spatial continuity of 0.92. Although the computation time (19.8 seconds) is not as good as K-means (16.2 seconds), it has a significant advantage over the OPTICS algorithm (50.5 seconds). The improvement comparison table at the bottom further quantifies the percentage improvement of this technical solution compared to each algorithm. Compared with the K-means algorithm, the accuracy is improved by 17.5%, the silhouette coefficient is improved by 24.6%, and the spatial continuity is improved by 37.3%. Compared with the worst-performing OPTICS algorithm, the overall improvement is as high as 38.9%. On average, compared with all compared algorithms, this technical solution improves the accuracy by 19.6%, the silhouette coefficient by 29.6%, the spatial continuity by 30.1%, and the calculation time is reduced by 32.8%, for an overall improvement of 26.4%.

[0135] In an optional embodiment, a distributed storage network is constructed based on a hierarchical cascade blockchain, risk partition data is encrypted to generate an early warning information package, a comprehensive matching degree is calculated through a double-layer spatial index to screen target early warning terminals, and the early warning information package is pushed in segments based on priority sorting, including:

[0136] Constructing a hierarchical and cascaded blockchain network, establishing a distributed storage structure, and generating early warning information blocks in the distributed storage structure;

[0137] Generate warning information based on risk partition data, asymmetrically encrypt the warning information to obtain warning information encrypted data, combine it with digital signature information to generate a warning information package, and store it in the warning information block;

[0138] Constructing a two-layer spatial index tree, calculating the comprehensive matching degree between the warning range in the warning information package and the warning terminal based on the multi-dimensional matching index, and screening the target warning terminal according to the comprehensive matching degree;

[0139] Mapping the risk level in the warning information package to a push priority to obtain priority data, and prioritizing the target warning terminals based on the priority data to obtain a warning terminal push sequence;

[0140] According to the early warning terminal push sequence, the early warning information package is fragmented and stored to obtain early warning information fragment data, and the early warning information package is pushed to the target early warning terminal through dynamic load balancing.

[0141] Construct a hierarchical and cascading blockchain network consisting of three layers: the root chain layer, the regional chain layer, and the terminal chain layer. The root chain layer is responsible for global data management and cross-regional data synchronization; the regional chain layer is responsible for data processing and storage within a specific geographic area; and the terminal chain layer directly interacts with the early warning terminal devices.

[0142] Based on this network, a distributed storage structure is established, using the IPFS (InterPlanetary File System) protocol to achieve distributed data storage. Each block contains a block header and a block body. The block header records information such as the timestamp, the hash value of the previous block, and the Merkle tree root hash, while the block body stores the warning information package data.

[0143] In practice, each node maintains a local ledger, ensuring data consistency through consensus algorithms such as the improved PBFT algorithm. For example, when a flood risk is detected, the system generates a warning information block on the blockchain network. This block contains information such as the warning ID, generation time, risk type, and impact range, and is linked to the previous block through a hash link.

[0144] The process of generating early warning information based on risk zoning data includes three steps: data collection, risk assessment and early warning information coding.

[0145] During the data collection phase, the system acquires real-time data from a multi-source sensor network (such as meteorological stations, hydrological monitoring stations, and earthquake monitoring stations) and analyzes it in combination with historical data. For example, in a typhoon warning scenario, the system collects data such as wind speed, air pressure, and rainfall to form a raw risk dataset.

[0146] During the risk assessment phase, the system processes and analyzes the collected data, identifying potential risks and determining their level. Risk levels are categorized into five levels: Very Low Risk (Level 1), Low Risk (Level 2), Medium Risk (Level 3), High Risk (Level 4), and Extremely High Risk (Level 5). For example, if rainfall in an area exceeds 200 mm within 24 hours and heavy rainfall is forecast for the next 12 hours, the system classifies the area as Level 4 (High Risk).

[0147] During the warning information coding stage, the system converts the risk assessment results into warning information in a standard format, including fields such as warning ID, release time, risk type, risk level, impact range (expressed in GeoJSON format), warning content and recommended measures.

[0148] The system encrypts warning information using asymmetric encryption algorithms (such as RSA-2048 or ECC) to ensure security during transmission. During encryption, the system uses the warning terminal's public key to encrypt the warning content. Only terminals holding the corresponding private key can decrypt and view the warning content. The system also generates a digital signature for the warning information using its own private key to verify its authenticity and integrity.

[0149] The system combines the encrypted warning information with the digital signature to generate a warning information package and stores it in the warning information block. The warning information package structure includes: a header (containing metadata such as the package ID, timestamp, and signature information) and a body (containing the encrypted warning information).

[0150] In order to efficiently match warning information with target terminals, the system constructs a two-layer spatial index tree, including a regional layer index and a terminal layer index.

[0151] The regional layer index adopts a quadtree structure to recursively divide the geographic space into grids of different granularities. It can be initially divided into large areas and then subdivided level by level. Each grid node stores the early warning terminal set and sub-grid information within the area.

[0152] The terminal-level index uses an R-tree structure, organized according to the terminal's geographic location and attribute information. Each leaf node in the R-tree represents an early warning terminal, containing its location coordinates and attribute information (such as terminal type, service population, and response capabilities). Non-leaf nodes represent the minimum bounding rectangle (MBR) of a group of terminals.

[0153] Based on the constructed two-layer spatial index, the comprehensive matching degree between the warning range in the warning information package and the warning terminal is calculated. The matching calculation takes into account multi-dimensional indicators and calculates the degree of overlap between the warning impact range and the terminal service range. For example, when a typhoon warning covers the coastal areas of Fujian, the system calculates the spatial intersection ratio between each terminal in the area and the warning range; and evaluates the matching degree between the terminal attributes and the characteristics of the warning information. For example, the attribute matching degree of school terminals for heavy rain warnings is higher than that for earthquake warnings. The terminal's processing ability for specific types of warnings is evaluated based on its historical response data.

[0154] The weighted sum of the matching degrees of the three dimensions above yields a comprehensive matching degree. The weights are: spatial matching degree 0.5, attribute matching degree 0.3, and historical response matching degree 0.2. When the comprehensive matching degree exceeds a preset threshold (e.g., 0.7), the terminal is selected as a target warning terminal.

[0155] The risk level in the warning information package is mapped to the push priority. The mapping relationship is: Level 5 corresponds to priority 1 (highest), Level 4 corresponds to priority 2, and so on. Based on the priority data, the system sorts the target warning terminals and generates a warning terminal push sequence.

[0156] For example, for a Level 5 typhoon warning, the system first maps it to priority 1. It then sorts the selected target warning terminals from highest to lowest comprehensive match, generating a push sequence. For terminals with the same comprehensive match, the system further considers factors such as the number of people served and critical infrastructure attributes.

[0157] Warning information packages are fragmented and stored based on the order in which they are pushed to warning terminals. This fragmentation strategy takes into account the size and complexity of the warning information, as well as the terminal's processing capabilities. For example, a flood warning containing a large amount of geographic information is divided into a basic information segment, a geographic data segment, and a detailed description segment, and pushed to terminals with different processing capabilities.

[0158] A dynamic load balancing mechanism is used to push warning packets, dynamically adjusting push strategies based on network conditions and terminal response times. During push notifications, the system monitors network bandwidth, node load, and transmission latency. If a node is overloaded (e.g., CPU utilization exceeding 85%), it automatically shifts some push tasks to less-loaded nodes.

[0159] Implement breakpoint resuming and push confirmation mechanisms to ensure reliable delivery of warning information. If a terminal fails to receive the information, the system automatically retries up to three times. For any terminals that still fail, the system logs the failure and attempts to push through an alternate channel.

[0160] Through the above method, the early warning information based on the hierarchical cascade blockchain is pushed efficiently, securely and accurately, which significantly improves the timeliness and coverage accuracy of disaster early warning.

[0161] In an optional embodiment, a two-layer spatial index tree is constructed, and based on a multi-dimensional matching index, a comprehensive matching degree between the warning range in the warning information package and the warning terminal is calculated. The target warning terminals obtained by screening according to the comprehensive matching degree include:

[0162] In the two-layer spatial index tree, the first layer constructs regional index nodes based on administrative regions, and the second layer constructs location index nodes based on geographic coordinates;

[0163] Divide the area into grid cells according to the warning range, and retrieve candidate warning terminals within the grid cells using a two-layer spatial index tree; construct a boundary polygon of the warning range, and calculate the distance from the candidate warning terminal to the boundary polygon to obtain the spatial distance matching degree;

[0164] Perform spatial calculation on the warning range and the administrative area boundary to obtain the overlapping area, and calculate the regional matching degree based on the overlapping area; extract the warning information reception delay and data integrity rate based on the historical warning data of the candidate warning terminal to obtain the terminal reliability index;

[0165] Constructing a traffic network topology structure, calculating the shortest path from the candidate warning terminal to the warning range to obtain path connectivity; obtaining population distribution data within the warning range, calculating the number of people covered by the candidate warning terminal to obtain a coverage index;

[0166] The spatial distance matching degree, the regional matching degree, the terminal reliability index, the path connectivity and the coverage index are normalized to determine a comprehensive score, and the warning terminal with the highest comprehensive score is selected as the target warning terminal.

[0167] In a specific embodiment, a two-layer spatial index tree is constructed, and the first layer constructs regional index nodes based on the administrative area division within the town, that is, the town is hierarchically organized according to grassroots administrative units such as communities and village committees. Each regional index node stores the boundary polygon coordinates, regional code, regional name, and a list of subordinate administrative units of its administrative area. The second layer constructs location index nodes based on geographic coordinates, organizes the early warning terminals within each grassroots administrative unit according to geographic coordinates, and uses a quadtree structure to subdivide the space. The location index node stores basic information such as the terminal's unique identifier, geographic coordinates, administrative unit code, and device type. For example, a town can be divided into multiple community or village committee-level regional index nodes, each community or village committee-level regional index node contains multiple grid unit-level regional index nodes, each grid unit-level regional index node contains multiple location index nodes, and each location index node corresponds to one or more early warning terminals.

[0168] The warning range gridding adopts an adaptive gridding method, dynamically adjusting the grid size according to the size and shape of the town warning range. For warning ranges less than 10 square kilometers, a uniform grid of 100 meters x 100 meters is used; for warning ranges between 10 and 50 square kilometers, a uniform grid of 200 meters x 200 meters is used; for warning ranges greater than 50 square kilometers, a uniform grid of 500 meters x 500 meters is used. After the gridding is completed, a two-layer spatial index tree is used to quickly retrieve candidate warning terminals within each grid cell. In specific implementation, the regional index node is first used to locate the grassroots administrative unit involved in the warning range, and then the location index node is used to locate the warning terminal within the grid cell.

[0169] The spatial distance matching degree is calculated. A boundary polygon for the warning range within the town is constructed, and the shortest distance from the candidate warning terminal to the boundary polygon is calculated. When the warning terminal is within the warning range, the distance is 0; when the warning terminal is outside the warning range, the shortest Euclidean distance from the warning terminal to the boundary polygon is calculated. The spatial distance matching degree is inversely proportional to the distance; the shorter the distance, the higher the matching degree. In practice, for terminals within the town warning range, the spatial distance matching degree is assigned a value of 1. For terminals outside the warning range, the matching degree is calculated using a distance decay function. For example, if a warning terminal is 500 meters from the warning range boundary, its spatial distance matching degree can be calculated as 0.85.

[0170] To calculate the regional matching degree, spatial calculations are performed between the warning range within the town and the boundaries of the grassroots administrative units to obtain the overlapping area. The regional matching degree is then calculated based on the overlapping area. The regional matching degree is equal to the ratio of the area of the overlapping area to the area of the grassroots administrative unit. For example, if the warning range covers 75% of a community, the regional matching degree of that community is 0.75. For grassroots administrative units that are completely within the warning range, the regional matching degree is 1; for grassroots administrative units that do not overlap with the warning range, the regional matching degree is 0.

[0171] The terminal reliability index is calculated based on historical warning data, primarily examining two aspects: the delay in receiving warning information and the data integrity rate. The delay in receiving warning information refers to the time difference between the issuance of a meteorological disaster warning in a town and its completion at the terminal; the shorter the delay, the better. The data integrity rate refers to the degree of completeness of the warning information successfully received by the terminal; the higher the integrity rate, the better. The terminal reliability index is calculated by collecting historical data on warning information received by the terminal within the town over the past three months and calculating the average delay and average integrity rate. For example, if the average delay for a warning terminal over the past three months was 0.8 seconds and the average integrity rate was 0.96, the terminal reliability index can be calculated as 0.92.

[0172] Path connectivity is calculated to construct the topology of the town's traffic network, consisting of road nodes and road edges. A road node represents a road intersection or endpoint, while a road edge represents the connection between two road nodes. Each road edge has attributes such as road grade, length, and capacity. The Dijkstra algorithm is used to calculate the shortest path from a candidate warning terminal to the warning area. Path connectivity is inversely proportional to the shortest path length; the shorter the path, the higher the connectivity. For example, if the shortest path from a warning terminal to the town's warning area is 2 kilometers long, the standardized path connectivity can be calculated as 0.9.

[0173] To calculate the coverage index, obtain population distribution data within the town's warning range and calculate the number of people covered by each candidate warning terminal. The coverage range of a warning terminal within a town depends on its device type. For example, a village broadcast terminal has a coverage radius of 500 meters, a community electronic display screen has a coverage radius of 150 meters, and a mobile app push notification has a coverage radius of a person's location. The coverage index is the ratio of the number of people covered by the terminal to the total population within the warning range. For example, if the total population within the warning range of a town is 5,000 and a warning terminal can cover 1,500 people, its coverage index is 0.3.

[0174] To calculate a comprehensive score, the spatial distance matching, regional matching, terminal reliability index, path connectivity, and coverage index are normalized and then weighted together to form a comprehensive score. The weights of each indicator can be adjusted based on the actual needs of the township, for example, a weight of 0.25 for spatial distance matching, 0.25 for regional matching, 0.25 for terminal reliability index, 0.1 for path connectivity, and 0.2 for coverage index. Ultimately, the warning terminal with the highest comprehensive score is selected as the target warning terminal. For example, in a township's rainstorm disaster warning scenario, if the comprehensive score of terminal A is 0.88 and the comprehensive score of terminal B is 0.76, terminal A is selected as the target warning terminal.

[0175] Existing methods for selecting meteorological disaster warning terminals in towns are mainly based on simple administrative area matching or fixed radius screening, which results in missed warnings and false alarms, and the terminal selection strategy cannot be dynamically adjusted according to actual conditions. For example, traditional methods send information to all terminals in the town during town warnings, or simply select terminals within a fixed radius around the warning point, without considering factors such as the terrain characteristics and population distribution within the town. The town-scale two-layer spatial index tree structure proposed in this application organically combines grassroots administrative units and geographic coordinate information, greatly improving the spatial retrieval efficiency within the town. At the same time, by introducing a multi-dimensional evaluation index system, it comprehensively considers factors such as spatial location, grassroots administrative units, terminal performance, town traffic connectivity and population coverage, making the selection of meteorological disaster warning terminals in towns more accurate. The starting point of the improvement of this application is to solve the problems of inaccurate terminal selection, incomplete coverage and low efficiency in the existing town warning system, and ultimately achieves the accurate push of meteorological disaster warning information in towns, reduces the waste of warning resources, improves warning efficiency, enhances the trust and acceptance of warning information by town residents, and provides strong support for meteorological disaster risk management in towns.

[0176] like Figure 3The figure shows a comparison of algorithm performance and resource consumption between this solution and traditional warning terminal selection methods. In terms of query time, this solution requires 125 milliseconds, which is slightly slower than the fixed radius search method's 87 milliseconds, but significantly better than the administrative region matching method's 213 milliseconds and the nearest neighbor search method's 189 milliseconds. In terms of memory consumption, this solution requires 178 MB, which is lower than the administrative region matching method's 245 MB, higher than the fixed radius search method's 123 MB and the nearest neighbor search method's 156 MB, placing it in the middle. However, when considering accuracy, this solution significantly outperforms other methods with a high accuracy of 92.5%. The next highest accuracy is only 77.8% for the nearest neighbor search method, 71.3% for the administrative region matching method, and a mere 58.7% for the fixed radius search method. In terms of CPU utilization, this solution's CPU utilization is 35.2%, lower than the administrative region matching method's 42.8% and the nearest neighbor search method's 38.7%, and only higher than the fixed radius search method's 21.5%. This data clearly demonstrates that while the fixed-radius search method performs best in terms of query speed and resource consumption, its accuracy is significantly lower. This technical solution, on the other hand, achieves an optimal balance of performance and accuracy while maintaining good query speed and moderate resource consumption. This fully demonstrates the comprehensive advantages of this technical solution in practical application scenarios, especially in critical scenarios requiring high-precision early warning terminal selection.

[0177] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent assessment and push of meteorological disaster risk classification for towns, characterized by: include: Obtain meteorological monitoring data from multiple monitoring points within the town and generate a meteorological monitoring data set; Based on the meteorological monitoring dataset, the spatiotemporal features are extracted through the spatiotemporal attention mechanism. The spatiotemporal features are adaptively fused with the preset geographic element features based on the dynamic weight mechanism of scene perception to obtain the fused feature vector. The fused feature vector is processed through a bidirectional long short-term memory network to generate the meteorological disaster risk level; Obtain the distribution information of each functional area in the town, generate a regional distribution data set, build a town vulnerability assessment model, and calculate the regional vulnerability assessment index; Based on multi-level regional vulnerability assessment indicators, a graph neural network and a time-series adaptive fusion network are constructed to dynamically integrate the meteorological disaster risk level with the regional vulnerability characteristics to obtain a comprehensive risk index. Spatial constraint clustering optimization is performed on the town area to generate risk zoning data. A distributed storage network is built based on a hierarchical cascade blockchain. The risk partition data is encrypted to generate an early warning information package. The target early warning terminal is screened by calculating the comprehensive matching degree through a double-layer spatial index. The early warning information package is pushed in segments based on priority sorting.

2. The method according to claim 1, characterized in that Based on the meteorological monitoring dataset, the spatiotemporal features are extracted through the spatiotemporal attention mechanism. The spatiotemporal features are adaptively fused with the preset geographic element features based on the dynamic weight mechanism of scene perception to obtain the fused feature vector. The fused feature vector is processed through a bidirectional long short-term memory network to generate meteorological disaster risk levels including: Construct a spatiotemporal attention module consisting of a temporal attention submodule and a spatial attention submodule, extract features from the meteorological monitoring dataset to obtain temporal features and spatial features, and perform weighted fusion calculation on the temporal features and spatial features to obtain spatiotemporal fusion features; Performing coding analysis on the current meteorological scene to generate a scene vector; calculating a dynamic fusion weight of the spatiotemporal fusion feature and a preset geographic element feature based on the scene vector, performing feature adaptive selection according to the dynamic fusion weight, and obtaining a fusion feature vector; The fused feature vector is input into the forward unit and the reverse unit of the bidirectional long short-term memory network, the forward hidden state and the reverse hidden state are determined and spliced, the risk level probability distribution is obtained by calculating the output weight matrix, and the meteorological disaster risk level is generated.

3. The method according to claim 2, characterized in that Performing coding analysis on the current meteorological scene to generate a scene vector; calculating the dynamic fusion weight of the spatiotemporal fusion feature and the preset geographic element feature based on the scene vector, performing feature adaptive selection according to the dynamic fusion weight, and obtaining the fusion feature vector includes: Constructing a dynamic scene-aware feature fusion network, comprising a scene encoder and an adaptive feature selector. The scene encoder employs a multi-head self-attention mechanism to interactively model the multidimensional meteorological elements of the current meteorological scene, generating a scene feature matrix. The scene feature matrix is then optimized through residual connections and normalization layers to generate a scene vector. Based on the scene vector, a feature importance evaluation module, a feature relevance calculation unit, and a scene adaptability calculation unit are constructed; wherein the feature relevance calculation unit calculates the mutual information score between the spatiotemporal fusion feature and the preset geographic element feature, and the scene adaptability calculation unit calculates the contribution score of each feature to the current scene; the mutual information score and the contribution score are weightedly combined to obtain a feature importance score; The adaptive feature selector constructs a soft gating mechanism based on the feature importance score, maps the feature importance score to the feature selection probability through the sigmoid function, dynamically adjusts the feature fusion weight according to the feature selection probability, realizes adaptive selection and fusion of features, and generates a fused feature vector.

4. The method according to claim 1, wherein Based on the multi-level regional vulnerability assessment indicators, a graph neural network and a time series adaptive fusion network are constructed to dynamically integrate the meteorological disaster risk level with the regional vulnerability characteristics to obtain a comprehensive risk index. Spatial constraint clustering optimization is performed on the town area to generate risk zoning data including: Acquiring regional vulnerability data at a current moment according to multi-level regional vulnerability assessment indicators, wherein the multi-level regional vulnerability assessment indicators include a population vulnerability indicator, an economic vulnerability indicator, an infrastructure vulnerability indicator, and an emergency response capability indicator; A graph neural network model is constructed, using each indicator in the regional vulnerability data as a graph node feature and constructing graph edge connections based on geographic adjacency. The spatial dependency features between graph node features are extracted through a graph convolutional layer, and the spatial dependency features at different scales are adaptively aggregated using an attention mechanism to obtain a regional vulnerability feature vector. A time series adaptive fusion network including a gated recurrent unit and a dynamic calibration module is constructed. The meteorological disaster risk level and the regional vulnerability feature vector are input into the gated recurrent unit to capture the time series fusion features. The dynamic calibration module adaptively calibrates and assesses the credibility of the time series fusion features based on historical data distribution to generate a comprehensive risk index with a confidence interval. The comprehensive risk index is input into the spatial constraint clustering algorithm to perform initial risk partitioning, spatial adjacency constraint items are introduced to ensure the spatial continuity of the initial risk partitions, and the boundaries of the initial risk partitions are optimized by the simulated annealing optimization algorithm to obtain risk partition data; The comprehensive risk index is input into the spatial constraint objective function for iterative optimization to obtain the initial risk partition. The local density value and the nearest high-density distance value are calculated based on the regional adjacency graph to identify the density peak point. The initial risk partition boundary is optimized with the density peak point to obtain the risk partition data.

5. The method according to claim 4, characterized in that The comprehensive risk index is input into the spatial constraint objective function for iterative optimization to obtain the initial risk partition. The local density value and the nearest high-density distance value are calculated based on the regional adjacency graph to identify the density peak point. The initial risk partition boundary is optimized with the density peak point. The risk partition data obtained includes: Receive a comprehensive risk index, including risk values of multiple risk sample points; The sum of the Euclidean distances from multiple risk sample points to the category center is calculated to obtain the intra-class distance. A spatial adjacency relationship is established based on the geographical locations of the multiple risk sample points, and the category difference values between the adjacent points are calculated to obtain a spatial adjacency constraint. The intra-class distance and the spatial adjacency constraint are combined to construct a spatial constraint objective function. The comprehensive risk index is input into the spatial constraint objective function, and the category center point is obtained through iterative optimization. Based on the category center point, spatial constraint clustering calculation is performed on multiple risk sample points to obtain the initial risk partition result; Constructing a regional adjacency graph based on the initial risk partitioning result, calculating the number of adjacent points within a preset cutoff distance for each risk sample point to obtain a first local density value set, selecting a target risk sample point from the first local density value set, determining adjacent risk sample points that are adjacent to the target risk sample point and have a local density value greater than the target risk sample point, calculating the minimum distance value from the target risk sample point to the adjacent risk sample point, and obtaining a nearest high-density distance value; A density peak point in the initial risk partition result is identified based on the local density value and the nearest high density distance value, and a boundary optimization is performed on the initial risk partition result based on the density peak point to obtain risk partition data.

6. The method according to claim 1, characterized in that Based on the hierarchical cascade blockchain, a distributed storage network is built to encrypt risk partition data to generate early warning information packages. The target early warning terminals are screened through the comprehensive matching degree calculated by the double-layer spatial index. The early warning information packages are pushed in fragments based on priority sorting, including: Constructing a hierarchical and cascaded blockchain network, establishing a distributed storage structure, and generating early warning information blocks in the distributed storage structure; Generate warning information based on risk partition data, asymmetrically encrypt the warning information to obtain warning information encrypted data, combine it with digital signature information to generate a warning information package, and store it in the warning information block; Constructing a two-layer spatial index tree, calculating the comprehensive matching degree between the warning range in the warning information package and the warning terminal based on the multi-dimensional matching index, and screening the target warning terminal according to the comprehensive matching degree; Mapping the risk level in the warning information package to a push priority to obtain priority data, and prioritizing the target warning terminals based on the priority data to obtain a warning terminal push sequence; According to the early warning terminal push sequence, the early warning information package is fragmented and stored to obtain early warning information fragment data, and the early warning information package is pushed to the target early warning terminal through dynamic load balancing.

7. The method according to claim 6, characterized in that Construct a two-layer spatial index tree, calculate the comprehensive matching degree between the warning range in the warning information package and the warning terminal based on the multi-dimensional matching index, and select the target warning terminals according to the comprehensive matching degree, including: In the two-layer spatial index tree, the first layer constructs regional index nodes based on administrative regions, and the second layer constructs location index nodes based on geographic coordinates; Divide the area into grid cells according to the warning range, and retrieve candidate warning terminals within the grid cells using a two-layer spatial index tree; construct a boundary polygon of the warning range, and calculate the distance from the candidate warning terminal to the boundary polygon to obtain the spatial distance matching degree; Perform spatial calculation on the warning range and the administrative area boundary to obtain the overlapping area, and calculate the regional matching degree based on the overlapping area; extract the warning information reception delay and data integrity rate based on the historical warning data of the candidate warning terminal to obtain the terminal reliability index; Constructing a traffic network topology structure, calculating the shortest path from the candidate warning terminal to the warning range to obtain path connectivity; obtaining population distribution data within the warning range, calculating the number of people covered by the candidate warning terminal to obtain a coverage index; The spatial distance matching degree, the regional matching degree, the terminal reliability index, the path connectivity and the coverage index are normalized to determine a comprehensive score, and the warning terminal with the highest comprehensive score is selected as the target warning terminal.

Images