Earthquake disaster loss dynamic assessment method and system based on multi-source disaster data

Abstract

This invention discloses a method and system for dynamic assessment of earthquake disaster losses based on multi-source disaster data. The invention relates to the field of earthquake disaster loss assessment technology. The method includes the following steps: S1, acquiring and preprocessing multi-source raw earthquake disaster data, wherein the multi-source raw data includes seismic motion parameter data and multi-source disaster data; S2, based on the preprocessed multi-source data, constructing a coupled driving model of the earthquake influence field and the multi-source disaster data, and generating a regional disaster loss heat map. This method and system for dynamic assessment of earthquake disaster losses based on multi-source disaster data acquires multi-source earthquake disaster data and constructs a coupled driving model of the earthquake influence field and the multi-source disaster data. By comprehensively considering the influence of multiple factors on earthquake disaster losses, it overcomes the limitations of traditional methods that rely on a single data source, and can more comprehensively and accurately reflect the true situation of earthquake disaster losses, providing a more reliable basis for rescue decisions.

Description

Technical Field

[0001] This invention relates to the field of earthquake disaster loss assessment, and more specifically, to a method and system for dynamic assessment of earthquake disaster losses based on multi-source disaster data. Background Technology

[0002] Accurate and timely assessment of earthquake damage is crucial for developing effective relief strategies, allocating relief resources rationally, and planning post-disaster reconstruction.

[0003] Traditional earthquake disaster loss assessment methods often rely on a single data source, such as assessment based solely on ground motion parameters or considering only building damage. However, earthquake disasters have a wide-ranging impact, involving multiple aspects, and a single data source cannot comprehensively and accurately reflect the true extent of disaster losses. For example, relying solely on ground motion parameters makes it difficult to fully account for the impact of factors such as population density and infrastructure distribution on disaster losses; while considering only building damage data ignores indirect losses caused by disruptions to transportation networks and damage to power and communication facilities.

[0004] Furthermore, traditional assessment methods are mostly static, making it difficult to reflect the dynamic changes in earthquake disaster losses over time. After an earthquake, disaster losses are not static; as time goes on, factors such as the occurrence of secondary disasters and the implementation of rescue operations cause disaster losses to continuously evolve. Static assessment methods cannot capture these changes in a timely manner, thus failing to provide the latest and most accurate information to support rescue decisions.

[0005] Therefore, existing earthquake disaster loss assessment methods have significant shortcomings in terms of data comprehensiveness and assessment dynamism. There is an urgent need for a new method that can integrate multi-source disaster data and achieve dynamic assessment in order to improve the accuracy and timeliness of earthquake disaster loss assessment. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic assessment method and system for earthquake disaster losses based on multi-source disaster data. This method solves the problem that most traditional assessment methods are static and cannot reflect the dynamic changes of earthquake disaster losses over time. After an earthquake, the disaster losses are not static. As time goes by, factors such as the occurrence of secondary disasters and the implementation of rescue operations will cause the disaster losses to evolve continuously. Static assessment methods cannot capture these changes in time, thus failing to provide the latest and most accurate information support for rescue decisions and failing to meet the needs of use.

[0007] This invention achieves the above objective through the following technical solution: a method for dynamic assessment of earthquake disaster losses based on multi-source disaster data, the method comprising the following steps:

[0008] S1. Acquire multi-source raw data of earthquake disaster and perform preprocessing. The multi-source raw data includes seismic motion parameter data and multi-source disaster data.

[0009] S2. Based on the preprocessed multi-source data, construct a coupled driving model of earthquake influence field and multi-source disaster data to generate regional disaster heat map.

[0010] S3. Using the regional disaster damage heat map as a continuous learnable variable, introduce it into the spatiotemporal evolution modeling framework. Through the fusion of graph neural network and temporal attention mechanism, construct a spatiotemporal dynamic evolution model of the disaster damage level regional map.

[0011] S4. Based on the aforementioned spatiotemporal dynamic evolution model, continuously track the evolution trajectory of the disaster in real time, predict the regional loss distribution in the next few hours, and output dynamic assessment results.

[0012] Furthermore, in step S1, the seismic motion parameter data includes:

[0013] Peak ground acceleration, peak ground velocity, response spectrum acceleration, and multi-source disaster data include building damage data, population density data, infrastructure distribution data, transportation network data, and historical disaster loss data;

[0014] Building damage data includes building structure type, construction year, and damage level classification;

[0015] Population density data uses fixed-size grids as statistical units;

[0016] Infrastructure distribution data includes the locations and service areas of key facilities in transportation, electricity, communications, water and gas supply;

[0017] Traffic network data includes road classification, traffic status, and the distribution of bridges and tunnels;

[0018] Historical disaster loss data refers to statistical data on casualties, economic losses, and the extent of damage from earthquakes of the same type and in the same region within a specified period.

[0019] Furthermore, the preprocessing in step S1 includes:

[0020] Data cleaning and normalization;

[0021] During the data cleaning process, outliers are identified and removed using preset criteria. Missing values ​​are filled using corresponding spatiotemporal interpolation methods based on the data type. Kriging interpolation is used for spatially continuous data, inverse distance weighted interpolation is used for discrete point data, and linear interpolation combined with neighboring time similarity correction is used for time-series missing data.

[0022] Normalization processes map all data values ​​to the [0,1] interval using a preset expression, thus unifying the data units.

[0023] Furthermore, in step S2:

[0024] Based on seismic motion parameter data, an improved attenuation model incorporating regional geological condition correction coefficients is used to calculate the seismic impact intensity of each spatial unit within the region.

[0025] A weight allocation model for multi-source disaster data was constructed. The comprehensive weight of various types of disaster data was determined by combining the analytic hierarchy process (AHP) with the entropy weight method. The AHP was used to construct a hierarchical structure and experts were invited to conduct pairwise comparisons. After consistency verification, the subjective weights were obtained. Then, the objective weights were calculated based on the information entropy of various types of disaster data using the entropy weight method. Finally, the weighted summation method was used to integrate the subjective and objective weights to obtain the comprehensive weight.

[0026] Furthermore, in step S2:

[0027] Based on the comprehensive weight of earthquake impact intensity and various disaster data, a coupled driving model is constructed. The disaster damage level of each spatial unit is calculated through a preset expression. The disaster damage level is divided into 0-5 levels, which correspond to no loss, slight loss, relatively minor loss, moderate loss, severe loss, and extremely severe loss, respectively.

[0028] Establish disaster damage level threshold rules, and determine the threshold for classifying each disaster damage level based on historical earthquake disaster loss statistics and industry standards;

[0029] Heatmap visualization technology is used to generate regional disaster damage heatmaps, and a gradient color system is used to map the disaster damage level to ensure that the visualization results are intuitive and easy to identify.

[0030] Furthermore, in step S3:

[0031] The regional disaster heat map is spatially discretized, with a fixed-size grid as the basic spatial unit. Each spatial unit serves as a node in the graph structure, and edges are constructed based on the adjacency relationships between spatial units to form a spatial graph structure.

[0032] Each node's feature vector contains the disaster damage level, earthquake impact intensity, and normalized values ​​of various disaster data for that spatial unit.

[0033] Furthermore, in step S3:

[0034] A time series is constructed based on time series observation data. An observation time interval threshold is set, and the spatial map structure corresponding to the regional disaster damage heat map at different times is used as the input sample of the time series to form a spatiotemporal data pair.

[0035] A spatiotemporal evolution modeling framework integrating graph neural networks and temporal attention mechanisms is constructed. This framework includes a spatial feature extraction layer, a temporal dependency capture layer, and a feature fusion layer.

[0036] Furthermore, the spatial feature extraction layer uses a graph convolutional neural network to extract features from the spatial graph structure at each time step and outputs a spatial feature vector.

[0037] The temporal dependency capture layer uses a temporal attention mechanism to assign weights to spatial feature vectors at different times, sets attention weight thresholds, and outputs temporal feature vectors.

[0038] The feature fusion layer uses a fully connected network to fuse spatial feature vectors and temporal feature vectors, outputting a spatiotemporal joint feature vector, and constructing a spatiotemporal dynamic evolution model based on this vector;

[0039] An adaptive moment estimator optimizer is used to minimize the loss function between the model's predicted values ​​and the actual observed values. A model training termination threshold is set to complete the model training.

[0040] Furthermore, in step S4:

[0041] Set a real-time data update threshold, input the real-time updated multi-source disaster data into the preprocessing module for processing, and generate real-time earthquake impact field data and disaster characteristic data;

[0042] Real-time data is input into the trained spatiotemporal dynamic evolution model, and iterative calculations are performed at preset time intervals to continuously output regional disaster heat maps at each time point and plot the disaster evolution trajectory curve.

[0043] Set the prediction time step rule and output the regional disaster loss heat map and corresponding loss distribution data at different time nodes in the next 1-6 hours;

[0044] Set a threshold for identifying high-risk areas and divide areas into different risk levels;

[0045] The results are visualized and dynamic evaluation reports are generated. These reports support dynamic updates and format export.

[0046] A dynamic earthquake disaster loss assessment system based on multi-source disaster data is provided for implementing the aforementioned dynamic earthquake disaster loss assessment method based on multi-source disaster data. The system includes:

[0047] Data acquisition module, data preprocessing module, coupled-driven model construction module, spatiotemporal dynamic evolution model construction module, dynamic evaluation module;

[0048] The data acquisition module is used to acquire multi-source raw data of earthquake disasters, including seismic motion parameter data and multi-source disaster data.

[0049] The data preprocessing module is used to clean, interpolate, and normalize multi-source raw data;

[0050] The coupled-driven model construction module is used to construct a coupled-driven model of earthquake influence field and multi-source disaster data based on preprocessed multi-source data, and generate regional disaster damage heat map.

[0051] The spatiotemporal dynamic evolution model construction module is used to introduce the regional disaster damage heat map as a continuous learnable variable into the spatiotemporal evolution modeling framework. Through the fusion of graph neural network and temporal attention mechanism, a spatiotemporal dynamic evolution model of the disaster damage level regional map is constructed.

[0052] The dynamic assessment module is used to continuously track the evolution trajectory of disasters in real time based on a spatiotemporal dynamic evolution model, predict the regional loss distribution in the next few hours, classify risk level areas, and output dynamic assessment results and visualization reports.

[0053] The beneficial effects of this invention are as follows:

[0054] 1. Acquire seismic motion parameter data and multi-source disaster data including building damage data, population density data, infrastructure distribution data, transportation network data, and historical disaster loss data. Construct a coupled driving model of earthquake impact field and multi-source disaster data. By comprehensively considering the impact of multiple factors on earthquake disaster losses, it overcomes the limitations of traditional methods that rely on a single data source. It can more comprehensively and accurately reflect the true situation of earthquake disaster losses and provide a more reliable basis for rescue decisions.

[0055] 2. By using regional disaster damage heat maps as continuous learnable variables and introducing a spatiotemporal evolution modeling framework, a spatiotemporal dynamic evolution model is constructed through the fusion of graph neural networks and temporal attention mechanisms. This model can continuously track the evolution trajectory of the disaster in real time and predict the regional loss distribution in the next few hours, realizing dynamic assessment of earthquake disaster losses. Compared with traditional static assessment methods, this invention can capture the changes in earthquake disaster losses over time in a timely manner, providing strong support for the dynamic adjustment of rescue operations and improving rescue efficiency and effectiveness.

[0056] 3. In the data preprocessing stage, multiple methods were used to clean, interpolate, and normalize the multi-source raw data. Outliers were identified and removed through preset criteria. Missing values ​​were filled in using appropriate spatiotemporal interpolation methods according to the data type. Inverse distance weighted interpolation was used for discrete point data, and linear interpolation combined with the method of correction for missing time series data was used. Normalization process mapped all data values ​​to the interval [0,1], and unified the data units. The preprocessing measures effectively improved the quality and consistency of the data, and provided a good data foundation for subsequent model construction and evaluation analysis.

[0057] 4. In constructing the weight allocation model for multi-source disaster data, the analytic hierarchy process (AHP) combined with the entropy weight method is used to determine the comprehensive weight of various types of disaster data. The AHP constructs a hierarchical structure and invites experts to conduct pairwise comparisons. After consistency testing, subjective weights are obtained. The entropy weight method calculates objective weights based on the information entropy of various types of disaster data. Finally, the weighted summation method is used to integrate subjective and objective weights to obtain the comprehensive weight. The comprehensive weight allocation method fully considers subjective experience and objective data information, making the weight determination more scientific and reasonable, and further improving the accuracy and reliability of the coupled-driven model.

[0058] 5. A regional disaster damage heat map is generated using heat map visualization technology, and a gradient color system is used to map the disaster damage level to ensure that the visualization results are intuitive and easy to identify. At the same time, the relevant results are visualized and a dynamic assessment report is generated. The report supports dynamic updates and format export. This visualization method enables decision-makers to intuitively and clearly understand the distribution and evolution trend of earthquake disaster losses, which facilitates rapid and scientific decision-making. Attached Figure Description

[0059] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0060] Figure 1 This is a flowchart illustrating the overall method of the present invention;

[0061] Figure 2 This is a flowchart of the data preprocessing process of the present invention;

[0062] Figure 3 This is a flowchart illustrating the construction process of the coupled-driven model of the present invention.

[0063] Figure 4 This is a flowchart illustrating the construction process of the spatiotemporal dynamic evolution model of the present invention.

[0064] Figure 5 This is a system block diagram of the present invention. Detailed Implementation

[0065] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0066] Example 1:

[0067] Please see Figure 1-4This invention provides a technical solution: a method for dynamic assessment of earthquake disaster losses based on multi-source disaster data, the method comprising:

[0068] S1. Acquire multi-source raw data of earthquake disaster and perform preprocessing. The multi-source raw data includes seismic motion parameter data and multi-source disaster data.

[0069] Among them, multi-source raw data refers to data sets from different channels and of different types. In this method, it includes seismic motion parameter data and multi-source disaster data. Seismic motion parameter data reflects the ground motion characteristics during an earthquake, such as the amplitude and frequency of seismic waves. Multi-source disaster data covers information on various disasters caused by earthquakes, which may include building damage, casualties, and infrastructure damage. These data come from a wide range of sources, such as on-site monitoring equipment, satellite remote sensing, social media, and news reports. Preprocessing involves a series of processing operations on the acquired multi-source raw data to improve data quality and make it more suitable for subsequent analysis and modeling. Preprocessing operations may include data cleaning, data normalization, and handling of missing data values.

[0070] S2. Based on the preprocessed multi-source data, construct a coupled driving model of earthquake influence field and multi-source disaster data to generate regional disaster heat map.

[0071] Among them, the earthquake impact field describes the degree and extent of the earthquake's impact on different regions. It comprehensively considers factors such as earthquake magnitude, epicenter location, and focal depth, and simulates the spatial distribution of earthquake energy through certain mathematical models and algorithms, thereby determining the degree of earthquake impact on different regions, such as which regions experienced stronger shaking and which experienced relatively weaker shaking. The coupled-driven model is a modeling method that correlates and interacts two or more different types of data or factors to jointly drive the operation of a model. In this method, earthquake impact field data is combined with multi-source disaster data... By coupling the data and analyzing their inherent connections and mutual influences, a model is constructed that can comprehensively reflect the disaster losses caused by earthquakes in a region. The regional disaster loss heat map visually displays the degree of disaster loss in different regions in the form of a heat map. In the heat map, different colors or shades of color represent different levels of disaster loss. The darker the color or the more it leans towards the color indicating high disaster loss, the more severe the disaster loss in that region is. Conversely, the lighter the color or the more it leans towards the color indicating low disaster loss, the less severe the disaster loss in that region is. In this way, the spatial distribution of earthquake disasters can be clearly seen.

[0072] S3. Using the regional disaster damage heat map as a continuous learnable variable, a spatiotemporal evolution modeling framework is introduced. Through the fusion of graph neural network and temporal attention mechanism, a spatiotemporal dynamic evolution model of the disaster damage level regional map is constructed.

[0073] Among them, continuously learnable variables are variables whose parameters can be continuously adjusted during the modeling process to better fit the data. Regional disaster damage heatmaps, as continuously learnable variables, mean that their parameters will be continuously optimized based on the input data and the model's learning objectives when constructing a spatiotemporal dynamic evolution model, thus more accurately reflecting the dynamic changes in regional disaster damage. The spatiotemporal evolution modeling framework is a modeling method used to describe and analyze the changing patterns of things in both time and space dimensions. It considers the state of things at different points in time and the differences in different spatial locations. By establishing corresponding mathematical models, it can capture the evolutionary trends and patterns of things in the spatiotemporal dimensions. In this method, it is used to study the dynamic changes of earthquake disaster losses in time and space. Graph neural networks are neural network models specifically designed for processing graph-structured data. Graph-structured data consists of nodes and edges; nodes represent entities, and edges represent the relationships between entities. In earthquake disaster loss assessment, different regions can be regarded as nodes, and the relationships between regions as edges. Graph neural networks can learn complex relationships and features between nodes, thus enabling better analysis and prediction of regional disaster damage. Temporal attention mechanisms, an extension of attention mechanisms specifically designed for processing time-series data, automatically learn the importance of data at different points in the time series. During modeling, different weights are assigned to data at different points in time based on their relevance to the current prediction task, making the model focus more on important data points and improving its modeling ability and prediction accuracy for time-series data. In dynamic assessment of earthquake disaster losses, temporal attention mechanisms can help the model better capture the changing patterns of disaster damage over time. The spatiotemporal dynamic evolution model of disaster damage level regional maps combines the advantages of graph neural networks and temporal attention mechanisms. Based on regional disaster damage heat maps, it is built within a spatiotemporal evolution modeling framework. This model can describe the dynamic changes of earthquake disaster losses in different regions and at different points in time, and predict future trends in disaster damage levels by learning patterns from historical data.

[0074] S4. Based on the spatiotemporal dynamic evolution model, continuously and in real time track the evolution trajectory of the disaster, predict the regional loss distribution in the next few hours, and output dynamic assessment results.

[0075] Among these, continuous real-time tracking of the disaster's evolution trajectory utilizes a pre-constructed spatiotemporal dynamic evolution model to continuously monitor and record the development and changes in earthquake disaster losses. By inputting new data in real time, the model can update its understanding of the disaster's evolution in a timely manner, depicting the spatial change path of the disaster over time, i.e., the disaster's evolution trajectory. Predicting the regional loss distribution in the next few hours involves analyzing and learning from historical and current data using the spatiotemporal dynamic evolution model to predict the disaster losses in different regions in the next few hours. The prediction results are given on a regional basis, providing the possible loss level for each region at a specific point in the future, thus forming the regional loss distribution for the next few hours. The dynamic assessment results are a comprehensive and dynamic evaluation of earthquake disaster losses obtained by integrating the continuously tracked disaster evolution trajectory and the predicted future regional loss distribution. These results can include the real-time disaster loss level of different regions, the trend of disaster loss changes over a period of time, and areas that may be severely affected, providing a scientific basis for disaster relief, resource allocation, and decision-making.

[0076] It should be noted that when using this system, acquiring and preprocessing multi-source raw data can comprehensively collect earthquake-related information. After processing, the data quality is improved, laying the foundation for accurate assessment. A coupled model of earthquake impact field and multi-source disaster data is constructed to generate regional disaster damage heat maps, which intuitively present the degree of disaster damage in different regions, facilitating a rapid understanding of the spatial distribution of disasters. By introducing the heat map as a continuous learnable variable into the spatiotemporal evolution framework and integrating graph neural networks and temporal attention mechanisms to construct the model, it can capture the dynamic changes of disaster damage in the spatiotemporal dimension, improving the accuracy and foresight of the assessment. Based on this model, the system can continuously track the disaster trajectory in real time, predict future loss distribution, and output dynamic results, providing timely and scientific decision-making basis for disaster relief and resource allocation, and minimizing earthquake disaster losses.

[0077] In one embodiment, acquiring multi-source raw data on earthquake disasters and performing preprocessing includes:

[0078] Acquire multi-source raw data, including seismic motion parameter data such as peak ground acceleration, peak ground velocity, and response spectrum acceleration, as well as multi-source disaster data such as building damage data, population density data, infrastructure distribution data, transportation network data, and historical disaster loss data;

[0079] Building damage data includes building structure type, construction year, and damage level classification;

[0080] Population density data are presented in 1km×1km grid units.

[0081] Infrastructure distribution data includes the locations and service areas of key facilities such as transportation, electricity, communications, water and gas supply;

[0082] Traffic network data includes road classification, traffic conditions, and the distribution of bridges and tunnels;

[0083] Historical disaster loss data includes statistics on casualties, economic losses, and disaster extent of similar earthquakes in the same region over the past 30 years.

[0084] Data cleaning was performed on the multi-source raw data to remove outliers and fill in missing values: the 3σ criterion was used to identify outliers, that is, when the deviation of a data sample from the mean of the data type exceeds 3 times the standard deviation, it is judged as an outlier and is removed; for missing values, the corresponding spatiotemporal interpolation method was used to fill in the missing values ​​according to the data type, namely, Kriging interpolation was used for spatially continuous data, inverse distance weighted interpolation was used for discrete point data, and linear interpolation combined with the method of nearby time similarity correction was used to fill in the missing values ​​for temporal series data.

[0085] The cleaned multi-source data is normalized to unify data units and eliminate the impact of differences in the numerical ranges of different indicators on the model. The normalization expression is as follows:

[0086]

[0087] in, Indicates the first The first class of data Each sample value Indicates the first Minimum value of class data, Indicates the first The maximum value of the class data, This represents the normalized sample value, with all data values ​​mapped to the [0,1] interval after normalization.

[0088] This design, which acquires and preprocesses multi-source raw earthquake data, covers various ground motion parameters and multi-source disaster data. It clarifies the specific content of various data such as building damage. The raw data is first cleaned, outliers are removed using the 3σ criterion, missing values ​​are filled using different spatiotemporal interpolation methods according to data type, and then normalized. This comprehensive collection of earthquake-related data lays the foundation for accurate assessment. Data cleaning improves data quality and avoids interference from outliers and missing values. Normalization unifies the units of measurement and eliminates the impact of differences in the numerical range of different indicators on the model, making the data more comparable. This allows the subsequent model to analyze the data more accurately, improving the reliability and accuracy of the assessment results and providing high-quality data support for earthquake disaster loss assessment.

[0089] In one embodiment, based on preprocessed multi-source data, a coupled driving model of earthquake influence field and multi-source disaster data is constructed to generate a regional disaster damage heat map, including:

[0090] Based on seismic motion parameter data, an improved attenuation model is used to calculate the seismic impact intensity of each spatial unit within the region. The improved attenuation model incorporates regional geological condition correction factors, and its expression is as follows:

[0091]

[0092] in, Representing spatial units The intensity of the earthquake's impact, ranging from 0 to 10, with higher values ​​indicating a stronger impact. Indicates earthquake magnitude, Richter scale. Representing spatial units Distance to epicenter (unit: km) The model parameters were obtained by fitting similar global earthquake observation data over the past 50 years, with values ​​ranging from a∈[0.8,1.2], b∈[1.5,2.5], c∈[0.3,0.7], to d∈[0.01,0.05]. This is a correction factor for regional geological conditions, based on spatial units. The soil and rock types were determined as follows: k = 0.8-1.0 for hard rock areas, k = 1.0-1.2 for semi-hard soil areas, and k = 1.2-1.5 for soft soil areas. Let N be the random error term, which follows a normal distribution N(0, 0.01). 2 );

[0093] A weight allocation model for multi-source disaster data was constructed, and the comprehensive weight of various types of disaster data was determined by combining the analytic hierarchy process (AHP) with the entropy weight method. , Indicates the category of disaster data. The specific rules for determining this are as follows:

[0094] Subjective weights were determined using the analytic hierarchy process (AHP): a hierarchical structure was constructed, consisting of a target layer (regional disaster damage assessment), a criterion layer (various disaster data), and a scenario layer (assessment results). Ten to fifteen experts in earthquake disaster assessment were invited to conduct pairwise comparisons of the importance of various disaster data, constructing a judgment matrix. After passing a consistency test (consistency index CI < 0.1), eigenvectors were calculated to obtain the subjective weights. ;

[0095] Objective weights are determined using the entropy weight method: objective weights are calculated based on the information entropy of various disaster data. ,in , Indicates the first In the disaster data, the first The normalized values ​​of each sample. For sample size, objective weights ;

[0096] Integrating subjective and objective weights: The weighted summation method is used to calculate the overall weight. ,in This is the weighting balancing coefficient, which is adjusted according to the actual application scenario. The value range is 0.4-0.6, and the default value is 0.5.

[0097] Based on the combined weights of earthquake impact intensity and various disaster data, a coupled driving model is constructed to calculate the disaster damage level of each spatial unit. Disaster damage levels are divided into 0-5 levels: Level 0 - No loss; Level 1 - Minor loss; Level 2 - Relatively minor loss; Level 3 - Moderate loss; Level 4 - Severe loss; Level 5 - Extremely severe loss. The expression is:

[0098]

[0099] in, Indicates the number of disaster data categories. Representing spatial units The normalized value of the k-th type of disaster data. The coupling function represents the relationship between the k-th type of disaster data and the intensity of earthquake impact. Specific coupling relationships are designed for different data types; for example, the coupling function for building damage data is... The coupling function for population density data is , This represents the rounding function;

[0100] Rules for setting disaster damage level thresholds: Based on historical earthquake disaster loss statistics and industry standards, the thresholds for classifying each disaster damage level are determined as follows:

[0101] Level 0 ( ),

[0102] Level 1 ( ),

[0103] Level 2 ( ),

[0104] Level 3 ( ),

[0105] Level 4 ( ),

[0106] Level 5 ( );

[0107] Based on the disaster damage level and threshold classification of each spatial unit, a regional disaster damage heat map is generated using heat map visualization technology. The map uses a gradient color system to map the disaster damage level, where blue corresponds to level 0, light green to level 1, dark green to level 2, yellow to level 3, orange to level 4, and red to level 5. The color transition is smooth and the contrast is significant, ensuring that the visualization results are intuitive and easy to distinguish.

[0108] This design, based on preprocessed data, first uses an improved attenuation model combined with geological condition correction coefficients to calculate the intensity of earthquake impact. Then, the analytic hierarchy process (AHP) and entropy weight method are used to determine the comprehensive weights of disaster data, constructing a coupled-driven model to calculate the disaster damage level. Threshold rules are set, and finally, a heat map visualization is used to generate a regional disaster damage heat map. The improved attenuation model considers geological conditions, making the calculation of earthquake impact intensity more accurate. The comprehensive weight determination method combines subjective and objective factors, making it more scientific and reasonable. The coupled-driven model can comprehensively assess the disaster damage level by integrating earthquake impact and disaster data. The threshold rules make the classification clearer, and the heat map visualization intuitively displays the disaster damage distribution, facilitating a rapid understanding of the disaster situation and providing a clear and intuitive basis for rescue decisions.

[0109] In one embodiment, the regional disaster damage heat map is used as a continuously learnable variable and introduced into a spatiotemporal evolution modeling framework. Through the fusion of graph neural networks and temporal attention mechanisms, a spatiotemporal dynamic evolution model of the disaster damage level regional map is constructed, including:

[0110] The regional disaster damage heat map is spatially discretized, using a 1km×1km grid as the basic spatial unit, with each spatial unit serving as a node in the graph structure. Adjacency relationships between spatial units are defined by shared boundaries or distances ≤2km, and edges are constructed to form the spatial graph structure. The feature vector of each node has an n+2 dimension, where n is the number of disaster data categories. The feature vector specifically includes the disaster damage level, earthquake impact intensity, and normalized values ​​of various disaster data for that spatial unit. The feature vector expression is as follows:

[0111]

[0112] A time series is constructed based on time-series observation data. The observation time interval threshold is set at 15 minutes, which can be adjusted according to the data update frequency, with an adjustment range of 5-60 minutes. The spatial map structure corresponding to the regional disaster damage heat map at different times is used as the input sample for the time series, forming a spatiotemporal data pair. ,in Indicates time Spatial diagram structure, Indicates the observation time ( , , (the observation time interval) Indicates the total number of observations;

[0113] A spatiotemporal evolution modeling framework integrating graph neural networks and temporal attention mechanisms is constructed. The framework includes a spatial feature extraction layer, a temporal dependency capture layer, and a feature fusion layer. The functions and parameter settings of each layer are as follows:

[0114] The spatial feature extraction layer employs a Graph Convolutional Neural Network (GCN) to extract features from the spatial graph structure at each time step, capturing the dependencies between spatial units. The GCN layer has 2-4 hidden layers, with each hidden layer containing 64-256 neurons (64 in the first layer, doubling the number in subsequent layers). The ReLU activation function is used, and the output is a spatial feature vector. (128 dimensions), the expression is:

[0115]

[0116] in, The adjacency matrix represents the spatial graph structure, using a 0-1 matrix where adjacent nodes are represented by 1s and non-adjacent nodes by 0s. Indicates time The feature matrix of each node has a dimension of N×(n+2), where N is the number of spatial units;

[0117] The temporal dependency capture layer employs a temporal attention mechanism to assign weights to spatial feature vectors at different times, highlighting the feature contributions of critical moments, such as the periods of rapid change in disaster situation, like 1 hour or 3 hours after an earthquake. An attention weight threshold of 0.05 is set; features at times below this threshold are considered to have extremely low contribution and are ignored, resulting in the output temporal feature vector. (128 dimensions), the expression is:

[0118]

[0119]

[0120]

[0121] If the denominator is 0, then ,

[0122] in, Indicates time For time The original attention weights, This represents the attention weights after threshold filtering. The attention calculation function uses scaled dot product attention, and its expression is: ,in , , , The dimension of the feature vector;

[0123] The feature fusion layer uses a fully connected network to process spatial feature vectors. With time series feature vectors For fusion, the fully connected network has one hidden layer (128 neurons), uses the sigmoid function as the activation function, and outputs a spatiotemporal joint feature vector. (64-dimensional), and based on the spatiotemporal joint feature vector, a spatiotemporal dynamic evolution model of the disaster-damaged area map is constructed:

[0124]

[0125] in, (Dimensions 64×128) (64×128) represents the weight matrix, initialized using a Xavier uniform distribution. (Dimension 64×1) represents the bias term, initialized with a constant of 0.1. This represents the Sigmoid activation function;

[0126] Using the feature vectors corresponding to continuously learnable disaster-level area maps as the optimization objective, the Adam adaptive moment estimator is employed to minimize the loss function between the model's predicted values ​​and the actual observed values. The optimizer parameters are set as follows: the initial learning rate is 0.001, adjusted using a cosine annealing strategy, decaying to 0.95 of the current value every 100 epochs; the weight decay coefficient is 0.0001; and the batch size is 32. The loss function expression is as follows:

[0127]

[0128] in, Indicates the number of spatial units. Indicates time No. The actual disaster damage level of each spatial unit This indicates the disaster damage level predicted by the model. This represents the regularization coefficient, with a value of 0.001. This represents the L2 regularization term, used to prevent the model from overfitting;

[0129] Set a model training termination threshold: when the loss function value decreases by less than 1e-5 over 10 consecutive epochs, or when the total number of training epochs reaches 500, stop model training and save the current optimal model parameters.

[0130] This design discretizes the regional disaster damage heat map to construct a spatial graph structure, forms spatiotemporal data pairs based on time-series data, and builds a framework that integrates a graph neural network and a temporal attention mechanism. It consists of spatial feature extraction, temporal dependency capture, and feature fusion layers, with parameters and optimization objectives set for each layer. An optimizer is used to minimize the loss function to train the model. Spatial discretization and graph structure construction facilitate the analysis of spatial relationships. The fusion framework can simultaneously capture spatial correlations and temporal dependencies. The spatial feature extraction layer extracts spatial features, the temporal dependency capture layer highlights key moment features, and the feature fusion layer integrates both. The optimizer and loss function settings enable the model to better learn data patterns, improve prediction accuracy, and provide a reliable model for dynamic assessment.

[0131] In one embodiment, based on a spatiotemporal dynamic evolution model, the trajectory of disaster evolution is continuously tracked in real time, and the regional loss distribution in the next few hours is predicted, outputting dynamic assessment results, including:

[0132] Set a real-time data update threshold: the update frequency of multi-source disaster data is no less than 30 minutes / time, the seismic motion parameter data is updated synchronously according to the real-time observation data of the monitoring station, and the update delay is ≤5 minutes. Input the real-time updated multi-source disaster data into the preprocessing module, and perform cleaning, interpolation and normalization processing according to the above method to generate real-time earthquake influence field data and disaster characteristic data.

[0133] Real-time data is input into the trained spatiotemporal dynamic evolution model. Through iterative calculation of the model, it is iterated once every 30 minutes, and the regional disaster damage heat map at each time moment is continuously output. Based on the temporal change of the disaster damage level of each spatial unit in the map, the disaster evolution trajectory curve is drawn. With time as the horizontal axis and disaster damage level as the vertical axis, it is grouped and displayed by spatial unit to realize real-time tracking of the disaster evolution trajectory.

[0134] Set the prediction time step rule: Based on the model's prediction accuracy verification results, determine the future prediction duration to be 1-6 hours, divide the prediction interval with 1 hour as the time node, that is, output the prediction results for the next 1 hour, 2 hours, ..., 6 hours. The prediction time step is consistent with the real-time update time interval (30 minutes / step).

[0135] Based on the current and historical times, the disaster loss evolution characteristics of the last 24 hours are used to output regional disaster loss heat maps and corresponding loss distribution data at different time nodes in the next 1-6 hours through the model's time series prediction capability. The loss distribution data includes: disaster loss level and damaged area of ​​each spatial unit. The total area of ​​spatial units with disaster loss level ≥2 is statistically analyzed and the economic loss is estimated. The economic loss per unit area is calculated by multiplying the average economic loss per unit area of ​​the same disaster loss level in history by the damaged area. The average economic loss per unit area is obtained by fitting regional economic statistics data of the past 10 years with historical disaster loss data.

[0136] Set a threshold for high-risk areas: spatial units with a predicted disaster level of ≥4 are designated as high-risk areas, those with a level of ≥3 and <4 are designated as medium-risk areas, those with a level of ≥2 and <3 are designated as low-risk areas, and those with a level of <2 are designated as safe areas.

[0137] The system visualizes real-time tracking results, future prediction results, and risk level classification results, generating dynamic assessment reports. These reports include disaster evolution trend analysis (the rate of change in disaster damage level in the past 24 hours, the expansion / contraction speed of high-risk areas), high-risk area early warning (clearly defining the spatial scope of high-risk areas, the expected duration, and the number of people affected), and loss estimation summary (total damaged area, total economic loss, and estimated casualties at different time points). The reports support dynamic updates, updating every 30 minutes, and can be exported to PDF, Excel, and other formats.

[0138] This design sets a real-time data update threshold, processes the real-time data, inputs it into the model, continuously outputs heat maps to depict the evolution trajectory, sets prediction time step rules, outputs future loss distribution data, identifies high-risk areas, and visualizes the results to generate dynamic assessment reports. Real-time data updates ensure timely assessments and can quickly reflect changes in the disaster situation. The prediction function allows for advance knowledge of future losses, buying time for the allocation of rescue resources. Identifying high-risk areas helps in prioritizing rescue efforts. The visual report intuitively displays multifaceted information, supports dynamic updates and export, and facilitates decision-makers' ability to monitor the disaster situation at any time, make scientific and reasonable rescue decisions, and reduce earthquake disaster losses.

[0139] Example 2:

[0140] Please see Figure 5 A dynamic earthquake disaster loss assessment system based on multi-source disaster data is used to implement the aforementioned dynamic earthquake disaster loss assessment method based on multi-source disaster data. The system includes:

[0141] Data acquisition module, data preprocessing module, coupled-driven model construction module, spatiotemporal dynamic evolution model construction module, dynamic evaluation module;

[0142] The data acquisition module is used to acquire multi-source raw data of earthquake disasters, which includes seismic motion parameter data and multi-source disaster data.

[0143] The data preprocessing module is used to clean, interpolate, and normalize multi-source raw data;

[0144] The coupled-driven model construction module is used to construct a coupled-driven model of earthquake influence field and multi-source disaster data based on preprocessed multi-source data, and generate regional disaster damage heat map.

[0145] The spatiotemporal dynamic evolution model construction module is used to introduce the regional disaster damage heat map as a continuous learnable variable into the spatiotemporal evolution modeling framework. Through the fusion of graph neural network and temporal attention mechanism, a spatiotemporal dynamic evolution model of the disaster damage level regional map is constructed.

[0146] The dynamic assessment module is used to continuously track the evolution trajectory of disasters in real time based on a spatiotemporal dynamic evolution model, predict the regional loss distribution in the next few hours, classify risk level areas, and output dynamic assessment results and visualization reports.

[0147] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0148] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for dynamic assessment of earthquake disaster losses based on multi-source disaster data, characterized in that, The method includes the following steps: S1. Acquire multi-source raw data of earthquake disaster and perform preprocessing. The multi-source raw data includes seismic motion parameter data and multi-source disaster data. S2. Based on the preprocessed multi-source data, construct a coupled driving model of earthquake influence field and multi-source disaster data to generate regional disaster heat map. S3. Using the regional disaster damage heat map as a continuous learnable variable, a spatiotemporal evolution modeling framework is introduced. The regional disaster damage heat map is spatially discretized, with a fixed-size grid as the basic spatial unit. Each spatial unit serves as a node in the graph structure, and edges are constructed based on the adjacency relationships between spatial units to form a spatial graph structure. The feature vector of each node contains the disaster damage level, earthquake impact intensity, and normalized values ​​of various disaster data for that spatial unit. A time series is constructed based on temporal observation data, and an observation time interval threshold is set. The spatial graph structure corresponding to the regional disaster damage heat map at different times is used as the input sample of the time series to form spatiotemporal data pairs. A spatiotemporal evolution modeling framework integrating graph neural networks and temporal attention mechanisms is constructed. This framework includes a spatial feature extraction layer, a temporal dependency capture layer, and a feature fusion layer. A spatiotemporal dynamic evolution model of the disaster damage level regional map is constructed. S4. Based on the aforementioned spatiotemporal dynamic evolution model, continuously track the evolution trajectory of the disaster in real time, predict the regional loss distribution in the next few hours, and output dynamic assessment results.

2. The method for dynamic assessment of earthquake disaster losses based on multi-source disaster data according to claim 1, characterized in that, In step S1, the seismic motion parameter data includes: Peak ground acceleration, peak ground velocity, response spectrum acceleration, and multi-source disaster data include building damage data, population density data, infrastructure distribution data, transportation network data, and historical disaster loss data; Building damage data includes building structure type, construction year, and damage level classification; Population density data uses fixed-size grids as statistical units; Infrastructure distribution data includes the locations and service areas of key facilities in transportation, electricity, communications, water and gas supply; Traffic network data includes road classification, traffic status, and the distribution of bridges and tunnels; Historical disaster loss data refers to statistical data on casualties, economic losses, and the extent of damage from earthquakes of the same type and in the same region within a specified period.

3. The method for dynamic assessment of earthquake disaster losses based on multi-source disaster data according to claim 1, characterized in that, The preprocessing in step S1 includes: Data cleaning and normalization; During the data cleaning process, outliers are identified and removed using preset criteria. Missing values ​​are filled using corresponding spatiotemporal interpolation methods based on the data type. Kriging interpolation is used for spatially continuous data, inverse distance weighted interpolation is used for discrete point data, and linear interpolation combined with neighboring time similarity correction is used for time-series missing data. Normalization processes map all data values ​​to the [0,1] interval using a preset expression, thus unifying the data units.

4. The method for dynamic assessment of earthquake disaster losses based on multi-source disaster data according to claim 1, characterized in that, In step S2: Based on seismic motion parameter data, an improved attenuation model incorporating regional geological condition correction coefficients is used to calculate the seismic impact intensity of each spatial unit within the region. A weight allocation model for multi-source disaster data was constructed. The comprehensive weight of various types of disaster data was determined by combining the analytic hierarchy process (AHP) with the entropy weight method. The AHP was used to construct a hierarchical structure and experts were invited to conduct pairwise comparisons. After consistency verification, the subjective weights were obtained. Then, the objective weights were calculated based on the information entropy of various types of disaster data using the entropy weight method. Finally, the weighted summation method was used to integrate the subjective and objective weights to obtain the comprehensive weight.

5. The method for dynamic assessment of earthquake disaster losses based on multi-source disaster data according to claim 4, characterized in that, In step S2: Based on the comprehensive weight of earthquake impact intensity and various disaster data, a coupled driving model is constructed. The disaster damage level of each spatial unit is calculated through a preset expression. The disaster damage level is divided into 0-5 levels, which correspond to no loss, slight loss, relatively minor loss, moderate loss, severe loss, and extremely severe loss, respectively. Establish disaster damage level threshold rules, and determine the threshold for classifying each disaster damage level based on historical earthquake disaster loss statistics and industry standards; Heatmap visualization technology is used to generate regional disaster damage heatmaps, and a gradient color system is used to map the disaster damage level to ensure that the visualization results are intuitive and easy to identify.

6. The method for dynamic assessment of earthquake disaster losses based on multi-source disaster data according to claim 1, characterized in that: The spatial feature extraction layer uses a graph convolutional neural network to extract features from the spatial graph structure at each time step and outputs a spatial feature vector. The temporal dependency capture layer uses a temporal attention mechanism to assign weights to spatial feature vectors at different times, sets attention weight thresholds, and outputs temporal feature vectors. The feature fusion layer uses a fully connected network to fuse spatial feature vectors and temporal feature vectors, outputting a spatiotemporal joint feature vector, and constructing a spatiotemporal dynamic evolution model based on this vector; An adaptive moment estimator optimizer is used to minimize the loss function between the model's predicted values ​​and the actual observed values. A model training termination threshold is set to complete the model training.

7. The method for dynamic assessment of earthquake disaster losses based on multi-source disaster data according to claim 1, characterized in that, In step S4: Set a real-time data update threshold, input the real-time updated multi-source disaster data into the preprocessing module for processing, and generate real-time earthquake impact field data and disaster characteristic data; Real-time data is input into the trained spatiotemporal dynamic evolution model, and iterative calculations are performed at preset time intervals to continuously output regional disaster heat maps at each time point and plot the disaster evolution trajectory curve. Set the prediction time step rule and output the regional disaster loss heat map and corresponding loss distribution data at different time nodes in the next 1-6 hours; Set a threshold for identifying high-risk areas and divide areas into different risk levels; The results are visualized and dynamic evaluation reports are generated. These reports support dynamic updates and format export.

8. A dynamic assessment system for earthquake disaster losses based on multi-source disaster data, characterized in that, The system, applied to the earthquake disaster loss dynamic assessment method based on multi-source disaster data as described in any one of claims 1-7, comprises: Data acquisition module, data preprocessing module, coupled-driven model construction module, spatiotemporal dynamic evolution model construction module, dynamic evaluation module; The data acquisition module is used to acquire multi-source raw data of earthquake disasters, including seismic motion parameter data and multi-source disaster data. The data preprocessing module is used to clean, interpolate, and normalize multi-source raw data; The coupled-driven model construction module is used to construct a coupled-driven model of earthquake influence field and multi-source disaster data based on preprocessed multi-source data, and generate regional disaster damage heat map. The spatiotemporal dynamic evolution model construction module is used to introduce the regional disaster damage heat map as a continuous learnable variable into the spatiotemporal evolution modeling framework. Through the fusion of graph neural network and temporal attention mechanism, a spatiotemporal dynamic evolution model of the disaster damage level regional map is constructed. The dynamic assessment module is used to continuously track the evolution trajectory of disasters in real time based on a spatiotemporal dynamic evolution model, predict the regional loss distribution in the next few hours, classify risk level areas, and output dynamic assessment results and visualization reports.

Images