Building function loss prediction dynamic updating method fused with social media data

Abstract

The application discloses a building group function loss prediction dynamic updating method fusing social media data, comprising the following steps: constructing a disaster situation real-time monitoring module, collecting social media data related to disasters in real time, and extracting local waterlogging observation information in a research area; simulating the spatio-temporal evolution of building group function state probability distribution, dividing emergency response blocks, and generating block function loss data sets under multiple rainfall scenarios; constructing and training a situation awareness model, inferring the block function loss of all unobserved emergency response blocks according to the local waterlogging observation information; constructing and training a spatio-temporal prediction model, using the function loss of all emergency response blocks observed and inferred at different times to replace the prediction values of the spatio-temporal prediction model at the corresponding time, and outputting the prediction results of the updated future multiple time steps through iteration. The application solves the problem that the function loss of a building group cannot be predicted and dynamically updated by fusing social media data in the prior art.

Description

Technical Field

[0001] This invention relates to the field of disaster management and emergency response technology, specifically to a method for dynamically updating the forecast of functional loss of building complexes by integrating social media data. Background Technology

[0002] Floods are among the most destructive natural disasters globally. The frequent occurrence of extreme floods in recent years has further highlighted the urgency of flood risk management.

[0003] Despite the significant advantages of social media data in terms of real-time performance and acquisition cost, existing technologies still have significant shortcomings in applying social media data to the dynamic updating of building cluster functional loss forecasts. Specifically: 1) Currently, the application of social media data in flood emergency response is mainly concentrated at the disaster site; research on the building and social aspects focuses primarily on building damage identification, rescue request extraction, economic loss estimation, and sentiment analysis, but these remain largely static assessments and have not yet established a mechanism to integrate real-time observation information into future forecast updates. 2) Existing research on forecast updates that integrate social media data focuses only on predicting inundation scenarios and has not yet extended to the forecast updates of building cluster functional loss, and suffers from insufficient computational efficiency. 3) Due to the unstructured, multimodal, sparse, and uneven spatiotemporal distribution of social media data, existing dynamic update methods based on three-dimensional variational, online learning, and incremental learning are difficult to directly apply. Therefore, a methodological system that can integrate social media data during disasters and support the dynamic updating of building cluster functional loss forecasts has not yet been formed, severely restricting the practical value of such data in emergency response and the accuracy of forecast results.

[0004] Currently, there are still many challenges in using social media data to dynamically update the forecast of functional loss in building complexes: 1) Social media data is highly sparse in its spatiotemporal distribution, and how to achieve global inference based on local observations is an urgent problem to be solved. 2) Existing forecast update methods that integrate social media data are mainly used to update inundation scenarios, rely on probabilistic inference of model parameter uncertainties, and need to be coupled with computationally expensive two-dimensional hydrodynamic models. There is still a lack of methods for rapid forecast updates of functional loss in building complexes. Summary of the Invention

[0005] To overcome the shortcomings of the above technologies, this invention provides a method for dynamically updating the forecast of functional loss of building complexes by integrating social media data. This method can solve the problem that the existing technology has not yet achieved the dynamic updating of the forecast of functional loss of building complexes by integrating social media data, and provides an accurate, efficient and low-cost solution for the dynamic updating of the forecast of functional loss of building complexes under urban flooding disasters.

[0006] Terminology Explanation:

[0007] 1. DEM: Digital Elevation Model.

[0008] 2. SWMM: Storm Water Management Model.

[0009] 3. WCA2D: Water Cell Automata 2D, a two-dimensional hydrodynamic cellular automata model.

[0010] 4. ERZs: Emergency Response Zones.

[0011] 5. ZFL: Zone-level functionality loss.

[0012] 6. ReLU: Rectified Linear Unit.

[0013] 7. LSTM: Long Short-Term Memory.

[0014] The technical solution adopted by this invention to overcome its technical problems is:

[0015] A method for dynamically updating the prediction of functional loss of building complexes by integrating social media data includes:

[0016] S1. Construct a real-time disaster monitoring module to collect and process disaster-related social media data in real time and extract local waterlogging observation information within the study area;

[0017] S2. Prepare training data: Divide building types into four categories: high-rise, multi-story, villa and stacked villa. Simulate the spatiotemporal evolution of the probability distribution of functional status of building groups, divide emergency response areas, and generate functional loss datasets of areas under various rainfall scenarios.

[0018] S3. Construct and train a situational awareness model. Based on local flooding observation information, obtain the observation function loss value of the locally observed emergency response area at the corresponding time, and infer the function loss of all unobserved emergency response areas.

[0019] S4. Construct and train a spatiotemporal prediction model. The input includes historical observed rainfall values, forecasted rainfall values ​​for a single future time step, and historical functional loss values ​​for emergency response areas. The output is the predicted value of functional loss for emergency response areas for a single future time step. The prediction results for multiple future time steps are obtained through iterative forecasting.

[0020] S5. In practical applications, the local waterlogging observation information at each moment is obtained through the real-time disaster monitoring module. The observed emergency response areas and their observation function loss values ​​are determined. The function loss of the unobserved emergency response areas is inferred using the situational awareness model. The predicted values ​​of the corresponding spatiotemporal prediction model are replaced with the function losses of all emergency response areas observed and inferred at different times. The updated forecast results for multiple future time steps are then output.

[0021] Furthermore, in step S2, the training data includes the training data required for both the situational awareness model and the spatiotemporal prediction model. To obtain the training data, the required input data includes meteorological variables, flood-prone environmental variables, and building attribute-related variables. Among them, the meteorological variables include 24-hour rainfall data measured by rain gauges, and rainfall sequences with 24-hour cumulative rainfall exceeding preset values ​​are selected as the analysis objects. Flood-prone environmental variables include DEM data, slope aspect data, curvature data, slope data, topographic humidity index data, distance to the river centerline, and distance to the pipeline outlet. Building attribute-related variables include the outline, number of floors, floor area, and building type of each building in the building complex.

[0022] Furthermore, step S2 specifically includes:

[0023] S21. Based on meteorological variables and flood-prone environmental variables of the study area, urban waterlogging inundation simulation is carried out by coupling SWMM and WCA2D models to obtain the waterlogging evolution process.

[0024] S22. Using a probabilistic prediction method for building functional states, simulate the spatiotemporal evolution of the probability distribution of different functional states of a building complex during multiple rainfall events; the building functional states include five types: I, II, III, V, and VII, with larger values ​​indicating a more severe impact on the buildings;

[0025] S23. Define the building whose probability of being in functional state I at any time under any rainfall sequence is not the highest as a potentially affected building. Use the K-means spatial clustering method to divide all potentially affected buildings into multiple emergency response zones. Define the functional loss of a zone as the average of the sum of the probabilities of all potentially affected buildings in functional states III to VII within that zone. In this way, construct a dataset of functional loss of emergency response zones at different times under various rainfall scenarios.

[0026] S24. Based on the rainfall sequence, calculate the time history curve of functional loss for each emergency response area under each single rainfall event, and calculate the Pearson correlation coefficient matrix between the areas accordingly. Use the average Pearson correlation coefficient matrix under all rainfall events as a measure of functional loss correlation.

[0027] Furthermore, in step S3, the situational awareness model includes two multi-head graph attention layers and ReLU activation functions connected to each multi-head graph attention layer. Each multi-head graph attention layer is configured with several attention heads and multi-dimensional hidden layers. The output layer of the situational awareness model is a graph attention layer with a dimension of 1.

[0028] Furthermore, in step S3, training the situational awareness model specifically includes:

[0029] (1) The input variables of the situational awareness model are ,in, This represents the total number of emergency response zones, and 1 represents the dimension of the functional loss value. This represents the dimensions of node-related attributes, which include the proportion of the four building types. The dimensions of environmental variables are represented, including DEM, curvature, distance to the pipeline outlet, and distance to the river centerline. The observed emergency response area is defined as a region where potentially affected buildings exceeding a set proportion are located within the observed flood coverage area; the observed function loss value is calculated based on the observed water depth. For each unobserved emergency response area, the prior function loss value is calculated using an inverse distance weighting method based on all observed function loss values. The output of the situational awareness model is... ,express The prediction function loss value of all emergency response areas at any given time is used, and only the prediction results for the unobserved emergency response areas are retained.

[0030] (2) For all rainfall sequences included in the meteorological variables, divide them into training set, validation set and test set according to a preset ratio; extract data from the spatial distribution of functional loss values ​​of each time step generated by step S2 in the rainfall sequence of the training set, and concatenate them with the corresponding node-related attribute variables and environmental variables to construct a complete node feature vector; train the situational awareness model independently under a preset number of observation ratios, and calculate the training loss.

[0031] (3) Verify and test the situational awareness models trained under several preset observation ratios, and calculate the verification loss and test loss respectively.

[0032] Furthermore, in step (2), when training the situational awareness model, a random masking strategy is introduced to simulate the situation where only a portion of the emergency response area is observable in reality, specifically including:

[0033] (1) For the spatial distribution results of the functional loss values ​​of all emergency response area nodes at each time step generated by step S2 for the rainfall sequence contained in the training set, randomly select a preset observation ratio of emergency response area nodes, retain the simulated functional loss values, and perform masking on the remaining nodes, that is, mask the functional loss values ​​of the remaining nodes.

[0034] (2) Calculate the prior functional loss value for all masked nodes according to step (1), and use it to fill the feature vector of the corresponding masked area node. Then, concatenate the functional loss value of all emergency response area nodes with their corresponding node-related attribute variables and environmental variables to form the input of the situational awareness model.

[0035] (3) The mask nodes are predicted by the situational awareness model. After the prediction is completed, the loss is calculated only between the predicted results and the simulated values ​​of the masked area nodes by the mean square error loss function.

[0036] Further, in step S4, the spatiotemporal prediction model includes a rainfall feature extraction module, a spatiotemporal feature capture module, and an emergency response area functional loss prediction module. The rainfall feature extraction module serves as the input module for the spatiotemporal prediction module, processing rainfall data containing multiple historical time steps and one future time step, and concatenating the processed rainfall data with the emergency response area functional loss tensor of the corresponding historical time step along the time dimension. The spatiotemporal feature capture module processes time series data with a graph structure, enabling each graph node to fuse features from the corresponding spatiotemporal domain. The emergency response area functional loss prediction module outputs the emergency response area functional loss prediction result with the same number of time steps as the input future rainfall time steps.

[0037] Furthermore, the rainfall feature extraction module in the spatiotemporal prediction model includes two types of input data: one containing historical data and the other containing other data. A time step and the future F =Rainfall data tensor at 1 time step and corresponding history Functional loss tensor of emergency response area for 24 hours ,in, Integer and , This indicates the batch size used in training the spatiotemporal prediction model. This indicates that the input channel size is 1, corresponding to the feature dimension of functional loss in the emergency response area, and each time step contains... Functional loss value information for each emergency response zone node; the rainfall feature extraction module includes an LSTM module and a fully connected layer connected after the LSTM module, and the rainfall data tensor. The input is fed into an LSTM module and processed by a fully connected layer to obtain a tensor. ,Will and By concatenating the data along the time dimension, the final input tensor is formed. .

[0038] Furthermore, the spatiotemporal feature capture module includes a two-stage cascaded spatiotemporal convolutional module and an output layer. The two-stage cascaded spatiotemporal convolutional module is used to progressively extract deep spatiotemporal features from the input sequence, and the output layer is used for feature compression and output mapping, and outputs the functional loss prediction result for a future single time step. Each spatiotemporal convolutional module includes two gated temporal convolutional layers and a spatial graph convolutional layer located between the two gated temporal convolutional layers. The gated temporal convolutional layer is used to perform convolution operations on the time series signal in the time dimension and combines different activation functions to achieve dynamic feature modeling. It adopts one-dimensional causal convolution and introduces nonlinearity through gated linear units. The spatial graph convolution uses a Chebyshev multinomial approximation strategy for approximate calculation. The output layer includes two gated convolutional layers and one two-dimensional convolutional layer.

[0039] Furthermore, the emergency response area functional loss prediction module includes a fully connected layer with both input and output dimensions of 1. The tensor output by the spatiotemporal feature capture module is passed through the fully connected layer with both input and output dimensions of 1, and the second dimension is compressed to obtain the functional loss prediction result for the next time step. .

[0040] Furthermore, in step S4, training the spatiotemporal prediction model specifically includes:

[0041] (1) Divide all rainfall sequences included in the meteorological variables into training set, validation set and test set according to a preset ratio;

[0042] (2) For the rainfall sequence included in the training set and the spatial distribution of the functional loss values ​​of all emergency response area nodes at each time step generated by step S2, a sliding time window mechanism is used to construct training sample pairs. The input of each training sample pair includes historical data. Rainfall time-series data at the current time step and one future time step, as well as historical data. The functional loss values ​​of each emergency response area at each time step are output as the predicted functional loss values ​​of each emergency response area at the next time step.

[0043] (3) In Fill in the time step before the rainfall value. A zero value, and supplement The functional loss value of the emergency response area corresponding to each time step is zero, and the spatiotemporal prediction model predicts... The functional loss of each emergency response area at each time step is calculated using the mean square error loss function. Then, the sliding time window is advanced by one time step, and the above process is repeated. For each 24-hour rainfall sequence, 24 training sample pairs are generated. All rainfall sequences in the training set are traversed, and the average mean square error function of all generated sample pairs is used as the training loss of the spatiotemporal prediction model.

[0044] (4) Validate and test the trained spatiotemporal prediction model, and calculate the validation loss and test loss respectively.

[0045] Furthermore, in step S4, the prediction results for multiple future time steps are obtained through iterative forecasting, specifically including:

[0046] Let the rainfall sequence be denoted as When predicting the functional loss value at the first time step, fill in the gaps before the first rainfall value. There are zero values, denoted as And supplement this The functional loss value of the emergency response area at each time point. Both values ​​are zero; this is achieved by inputting the rainfall sequence. and the corresponding historical time of emergency response area functional loss value get Prediction results corresponding to time steps ;

[0047] When predicting the functional loss value at the second time step, the rainfall sequence used becomes , Will Prediction results corresponding to time steps As part of the input, namely, the historical moment-in-time emergency response area functional loss value used. ,get Prediction results corresponding to time steps ;

[0048] This process is repeated 24 times until the time-series prediction results of functional loss for all emergency response areas under the entire 24-hour rainfall sequence are generated.

[0049] The beneficial effects of this invention are as follows: 1. It constructs a multi-module collaborative intelligent update framework, improving the technical system for real-time perception and functional prediction of urban flooding disasters driven by social media data. 2. To address the limitation of sparse spatiotemporal distribution of observation information, it innovatively introduces graph attention networks and random mask training strategies, constructing a situational awareness model adaptable to different observation ratios, and realizing global inference of regional functional loss based on sparse local observation information. 3. To meet the timeliness requirements of emergency response, it establishes a spatiotemporal prediction model integrating spatiotemporal graph convolutional structures and long short-term memory networks, constructing an input mechanism based on prior disaster observations, and realizing efficient dynamic updates of building cluster functional loss and evacuation demand forecasts. 4. This invention dynamically couples real-time observation with prior knowledge of physical simulation, providing an auxiliary tool for urban flooding emergency response that combines accuracy and real-time performance. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the dynamic update method for predicting functional loss of building complexes by integrating social media data, as described in this invention.

[0051] Figure 2 This is a schematic diagram of the architecture of the situational awareness model described in an embodiment of the present invention.

[0052] Figure 3 This is a schematic diagram of the training strategy for the situational awareness model described in an embodiment of the present invention.

[0053] Figure 4 This is a schematic diagram of the spatiotemporal prediction model described in an embodiment of the present invention.

[0054] Figure 5 This is a schematic diagram of the training strategy for the spatiotemporal prediction model described in an embodiment of the present invention.

[0055] Figure 6 This is a schematic diagram of the spatial distribution of the emergency response zones according to an embodiment of the present invention.

[0056] Figure 7 This is a statistical chart showing the quantity distribution of the four building types within each emergency response zone as described in this embodiment of the invention.

[0057] Figure 8 This is the average Pearson correlation coefficient matrix between emergency response zones as described in this embodiment of the invention.

[0058] Figure 9 This is a schematic diagram illustrating the hourly forecast and measured rainfall and the amount of available social media data from 16:00 on September 5th to 16:00 on September 6th, as described in an embodiment of the present invention.

[0059] Figure 10This is a schematic diagram illustrating an application scenario of the dynamic update method for predicting functional loss of building complexes by integrating social media data, as described in an embodiment of the present invention.

[0060] Figure 11 This is a schematic diagram showing the water depth distribution of waterlogged points observed at different times according to an embodiment of the present invention, after local expansion.

[0061] Figure 12 This is a spatiotemporal distribution map of the functional loss value of the emergency response area at 23:00 on September 5th, as described in this embodiment of the invention, and its corresponding statistical histogram.

[0062] Figure 13 This is a spatiotemporal distribution map of the functional loss values ​​of the emergency response area at 00:00 on September 6, as described in this embodiment of the invention, and its corresponding statistical histogram.

[0063] Figure 14 This is a spatiotemporal distribution map of the functional loss value of the emergency response area at 01:00 on September 6, as described in this embodiment of the invention, and its corresponding statistical histogram.

[0064] Figure 15 This is a spatiotemporal distribution map of the functional loss value of the emergency response area at 02:00 on September 6, as described in this embodiment of the invention, and its corresponding statistical histogram.

[0065] Figure 16 This is a schematic diagram of the forecast update results after fusing social media data at different reporting times, as described in an embodiment of the present invention. Detailed Implementation

[0066] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings. These are merely exemplary embodiments of the present invention; however, it should be understood that the present invention can be implemented in various forms and is not limited to the embodiments described herein. These embodiments are provided to enable those skilled in the art to understand the present invention more clearly and thoroughly.

[0067] This invention discloses a method for dynamically updating the prediction of functional loss in building complexes by integrating social media data, such as... Figure 1 As shown, it includes the following steps:

[0068] S1. Construct a real-time disaster monitoring module to collect and process disaster-related social media data in real time and extract local waterlogging observation information within the study area;

[0069] S2. Prepare training data: Divide building types into four categories: high-rise, multi-story, villa and stacked villa. Simulate the spatiotemporal evolution of the probability distribution of functional status of building groups, divide emergency response areas, and generate functional loss datasets of areas under various rainfall scenarios.

[0070] S3. Construct and train a situational awareness model. Based on local flooding observation information, obtain the observation function loss value of the locally observed emergency response area at the corresponding time, and infer the function loss of all unobserved emergency response areas.

[0071] S4. Construct and train a spatiotemporal prediction model. The input includes historical observed rainfall values, forecasted rainfall values ​​for a single future time step, and historical functional loss values ​​for emergency response areas. The output is the predicted value of functional loss for emergency response areas for a single future time step. The prediction results for multiple future time steps are obtained through iterative forecasting.

[0072] S5. In practical applications, the local waterlogging observation information at each moment is obtained through the real-time disaster monitoring module. The observed emergency response areas and their observation function loss values ​​are determined. The function loss of the unobserved emergency response areas is inferred using the situational awareness model. The predicted values ​​of the corresponding spatiotemporal prediction model are replaced with the function losses of all emergency response areas observed and inferred at different times. The updated forecast results for multiple future time steps are then output.

[0073] Steps S1 to S5 are described in detail below.

[0074] S1. Construct a real-time disaster monitoring module to collect and process disaster-related social media data in real time and extract localized flooding observation information within the study area. Specifically, this includes the following:

[0075] (1) Disaster-related social media data are collected in real time through the disaster real-time monitoring module. The data acquisition channels include at least various social media platforms, information platforms, and news platforms. The retrieval process is based on the set keyword set and time range. Due to the differences in the webpage structure of different platforms, the data collectability varies. For some platforms, the data is automatically collected using a web crawler program developed based on the Python language Scrapy framework; while for other platforms, the data is manually filtered and downloaded by browsing.

[0076] (2) Process the collected raw multi-source social media data, including at least cleaning, deduplication and noise reduction operations, and then classify the disaster relevance to form structured social media data and store it. The structured social media data includes at least three types of multimodal data: text, pictures and videos.

[0077] (3) Fine-grained water depth information mining is performed on the obtained multimodal data, including geographic location information extraction, water depth information extraction and local area water depth expansion, so as to extract local waterlogging observation information in the study area.

[0078] Specifically, the extraction of geographic location information can be divided into two methods depending on the data source:

[0079] The first method is to directly extract structured location information: For text data that has its own latitude and longitude coordinates, its structured geographic fields can be directly extracted.

[0080] The second method is a geospatial deduction method based on image content: For image or video data with incomplete text descriptions but containing visual cues, extract iconic geographical elements from the image or video, such as landmark buildings, shop signs on both sides of the street, or attached text address information. Match these extracted iconic geographical elements with street view image data provided by virtual globe software to deduce the specific geographical location. Then, convert all extracted geographical locations into specific latitude and longitude coordinates.

[0081] Specifically, in the process of extracting water depth information, for images or videos showing flood scenes, a flood depth classification mapping rule applicable to this embodiment is formulated based on the level conversion rules between different reference objects and water depth. Based on the flood depth classification mapping rule, water depth is identified in at least the following ways: Image or video-based classification identification method: For flood scenes identified in images or videos, they are first divided into different severity levels, and then converted into corresponding quantitative water depth values. In this embodiment, three typical reference objects are selected: human body, shared bicycle, and car. Water depth values ​​corresponding to different flood severity levels are defined for each type of reference object, as shown in Table 1.

[0082] Table 1 Mapping Rules for Submerged Water Depth Classification

[0083]

[0084] Specifically, during the local water depth expansion process, after identifying floodwater accumulation points from social media data, it is assumed that the water surface elevation of the adjacent areas within a preset range around the floodwater accumulation point is consistent with the water surface elevation of the grid cell containing the floodwater accumulation point. The observed water depth information of the floodwater accumulation point is then expanded to the surrounding area to obtain the local flood water depth distribution. In this embodiment, at each observation time, a buffer zone with a radius of 300m is constructed for all floodwater accumulation points. Combined with DEM data, pixels within the buffer zone with an elevation difference of no more than 3m are identified as adjacent areas with the same floodwater surface elevation. Subsequently, based on the water depth and elevation information of the floodwater accumulation points and the elevation information of each neighboring pixel that meets the conditions, the water depth of all neighboring pixels that meet the conditions is calculated to obtain the local flood water depth distribution. To address the pixel conflict problem caused by the overlap of adjacent floodwater accumulation point buffer zones during the area expansion process, this embodiment adopts the maximum water depth value as a conservative estimation strategy to ensure the controllability of the risk in emergency response.

[0085] S2. Prepare training data: Divide building types into four categories: high-rise, multi-story, villa, and stacked villa. Simulate the spatiotemporal evolution of the probability distribution of functional status of building groups, divide emergency response areas, and generate functional loss datasets for various rainfall scenarios.

[0086] In this embodiment, the training data includes the training data required for both the situational awareness model and the spatiotemporal prediction model. To obtain the training data, the required input data includes meteorological variables, flood-prone environmental variables, and building attribute-related variables. Meteorological variables include 24-hour rainfall data measured by rain gauges. Preferably, in this embodiment, rainfall sequences with a 24-hour cumulative rainfall exceeding a preset value of 350 mm are selected as the analysis objects. Each rainfall sequence is considered a uniform rainfall event, meaning a rainfall event in which the hourly rainfall is spatially evenly distributed within the study area. Flood-prone environmental variables include DEM data, slope aspect data, curvature data, slope data, topographic humidity index data, distance to the river centerline, and distance to the pipeline outlet. Building attribute-related variables include the outline, number of floors, floor area, and building type of each building in the building complex.

[0087] Furthermore, step S2 specifically includes:

[0088] S21. Based on meteorological variables and flood-prone environmental variables of the study area, the SWMM and WCA2D models are coupled to simulate urban flooding and obtain the evolution process of urban flooding.

[0089] S22. The Building Functionality State Forecasting (BFS-Forecasting) method is used to forecast the probability of building functional states. Combined with meteorological variables, flood-prone environmental variables, and building attribute-related variables in the study area, the spatiotemporal evolution scenarios of the entire building complex in the study area under several rainfall events with different probability distributions of functional states are simulated. The BFS-Forecasting method is an existing method.

[0090] In this embodiment, the building functional states include five types: I, II, III, V, and VII. A higher value indicates a more severe impact on the building. Specifically, functional state I indicates that all residential units in the building can meet their various needs; functional state II indicates that the advanced needs of the bottom residential unit cannot be met, while the various needs of other residential units can be met, with advanced needs including at least those related to office work and cleaning; functional state III indicates that the basic needs of the bottom residential unit cannot be met, while the various needs of other residential units can be met, with basic needs including at least basic physiological needs such as food; functional state V indicates that the safety needs of the bottom residential unit cannot be met, while the various needs of other residential units can be met, with safety needs including at least life safety needs; and functional state VII indicates that the safety needs of all residential units in the building cannot be met. High-rise buildings refer to buildings with 7 or more floors, and multi-story buildings refer to buildings with 4 to 6 floors. Generally speaking, for high-rise and multi-story buildings, the bottom residential unit refers to the first-floor residents; for villas, the bottom residential unit is the entire residential unit, which may contain several floors; for stacked villas, the bottom residential unit refers to the lower stack, which is the first and second floors if the stack contains two floors.

[0091] Specifically, the BFS-Forecasting method considers the water depth resistance distribution of components such as electronic sockets, bottom switches, middle switches, refrigerators, kitchen countertops, water heaters, air conditioners, and computers. It uses Monte Carlo simulation to deduce the vulnerability curves of different building types corresponding to different functional states, thereby obtaining the probability distribution of the entire building complex in the study area at different times under several rainfall events in different functional states.

[0092] S23. If the probability of a building being in functional state I at any given moment under any rainfall sequence is not the highest among all possible states, then that building is defined as a potentially affected building. Then, the K-means spatial clustering method is used to divide all potentially affected buildings into multiple emergency response zones, denoted as ERZs. In the K-means-based spatial clustering process, the standard for determining the number of clusters is that the average number of buildings contained in each emergency response zone is between 10 and 15.

[0093] When an evacuation requirement arises when a building is in functional states III to VII, the functional loss of a region is defined as the average of the sum of probabilities of all potentially affected buildings in that region being in functional states III to VII, denoted by ZFL, and expressed as in formula (1). Based on this, a dataset of functional losses of emergency response regions at different times under various rainfall scenarios is constructed:

[0094] (1)

[0095] In formula (1), express Time of the first Functional loss value of each emergency response zone Indicates the first The number of potentially affected buildings in each emergency response zone This indicates the corresponding building index. Indicates the possible building function status categories, Indicates the first The building is in The time function status category is The probability value.

[0096] S24. Based on the rainfall sequence, calculate the time history curve of functional loss of each emergency response area under each single rainfall event, and calculate the Pearson correlation coefficient matrix between each emergency response area under the rainfall event. Then, take the average value of the Pearson correlation coefficient matrix obtained under all rainfall events, and use the average value as a measure of functional loss correlation between emergency response areas.

[0097] S3. Construct and train a situational awareness (SA) model. Based on local flooding observation information, obtain the observation function loss value of the locally observed emergency response area at the corresponding time, and infer the function loss of all unobserved emergency response areas.

[0098] In this embodiment, as Figure 2 As shown, the situational awareness model includes two multi-head graph attention layers and ReLU activation functions connected to each multi-head graph attention layer to enhance nonlinear expressive power. Each multi-head graph attention layer is configured with 16 attention heads and a 128-dimensional hidden layer. The output layer of the situational awareness model is a 1-dimensional graph attention layer. The adjacency matrix used by each multi-head graph attention layer and graph attention layer in the situational awareness model is determined as follows: Based on the average Pearson correlation coefficient matrix obtained in step S24, a correlation coefficient threshold is first set. , The value ranges from 0.7 to 1, and the correlation coefficient of functional loss between any two areas is... At that time, the adjacency matrix used in the situational awareness model Corresponding element in Conversely, then set .

[0099] In step S3 of this embodiment, as follows Figure 3 As shown, training the situational awareness model specifically includes:

[0100] (1) Each emergency response zone node contains three types of feature vectors, namely the functional loss value of the emergency response zone, node-related attribute variables and environmental variables. Among them, the node-related attribute variables include the proportion of the four building types in each emergency response zone, and the environmental variables include DEM, curvature, distance to the pipeline outlet and distance to the river centerline. Each indicator in the environmental variables is represented by the average value of the corresponding indicators at the location of the core of all buildings in each emergency response zone.

[0101] The input variables of the situational awareness model are It is composed of three types of features: the functional loss value of the emergency response area, node-related attribute variables, and environmental variables, which are spliced ​​together along the feature dimension. This represents the total number of emergency response zones, and 1 represents the dimension of the functional loss value. The dimensions representing the node's related attributes. Dimensions representing environment variables, such as Figure 2 As shown; the emergency response area is divided into observed emergency response areas and unobserved emergency response areas. The observed emergency response area is defined as an area where more than a certain percentage of potentially affected buildings are located within the observed flood coverage area. The preferred percentage here is 70%. The observed functional loss value is calculated based on the observed water depth using the BFS-Forecasting method and formula (1), and formula (1) only considers buildings located within the observed flood coverage area. For each unobserved emergency response area, the prior functional loss value is calculated based on all observed functional loss values ​​using the inverse distance weighting method, as shown in the following formula:

[0102] (2)

[0103] In formula (2), Indicates the first The prior functional loss value of each emergency response zone This represents the set of observed emergency response areas. Indicates the first The emergency response zone and the first The European distance between the centroids of each emergency response zone Indicates the first Functional loss values ​​obtained from observations in each emergency response zone The power-law weight parameter representing the distance is set to 1. The output of the situational awareness model is... , indicating in The predicted function loss value for all emergency response areas at any given time is used, and only the prediction results for the unobserved emergency response areas are retained; for the observed emergency response areas, the observation function loss value is directly retained.

[0104] (2) For all 24-hour rainfall sequences included in the meteorological variables, the training set, validation set, and test set are divided according to a preset ratio. During the training process, data are extracted sequentially from the spatial distribution of functional loss values ​​at each time step generated by step S2 in the rainfall sequence of the training set, and concatenated with the corresponding node-related attribute variables and environmental variables to construct a complete node feature vector. Then, the situational awareness model is trained independently under various preset observation ratios, and the training loss is calculated.

[0105] (3) Verify and test the situational awareness models trained under several preset observation ratios, and calculate the verification loss and test loss respectively.

[0106] In this preferred embodiment, 90% of the rainfall sequences are used for training and validation of the situational awareness model (and further divided into training and validation sets in a 7:3 ratio), while the remaining 10% of the rainfall sequences are used for model testing and performance evaluation. For each rainfall sequence used for model training, validation, and testing, the hourly functional loss distribution scenario of the emergency response area, simulated using the BFS-Forecasting method, is summarized.

[0107] Preferably, when training the situational awareness model, a random masking strategy is introduced to simulate the scenario where only a portion of the emergency response area is observable in reality, specifically including:

[0108] (1) For the spatial distribution results of the functional loss values ​​of all emergency response area nodes at each time step generated by step S2 for the rainfall sequence contained in the training set, a preset observation ratio of emergency response area nodes is randomly selected, the simulated functional loss values ​​are retained, and the remaining nodes are masked, that is, the functional loss values ​​of the remaining nodes are masked.

[0109] (2) Calculate the prior functional loss value for all masked nodes according to step (1) and use it to fill the feature vector of the corresponding masked area node. Then, concatenate the functional loss value of all emergency response area nodes with their corresponding node-related attribute variables and environmental variables to form the input of the situational awareness model.

[0110] (3) The mask nodes are predicted by the situational awareness model. After the prediction is completed, the loss is calculated by the mean square error loss function only for the difference between the predicted result and the simulated value of the masked area node.

[0111] To enhance the applicability of the situational awareness model under different observation ratios, this embodiment independently trains the model under five predefined observation ratios (10%, 20%, 30%, 40%, and 50%), thereby improving the model's robustness and prediction accuracy under varying information availability. Based on the situational awareness models trained under the above five preset observation ratios, scatter plots are generated using the model testing results and fitted to obtain an index to measure the goodness of fit. If the scatter plots obtained under various preset observation scales... If all values ​​are greater than 0.9, then the selected... The value meets the requirements; otherwise, Increase by 0.05 and repeat the above steps until the condition is met. The condition is met until all values ​​are greater than 0.9. Preferably, in this embodiment, [the following is a more detailed description:] Set it to 0.95.

[0112] The training process of the aforementioned situational awareness model was implemented in the Python environment using the PyTorch 2.0 framework. The mean squared error (MSE) loss function was chosen, and the batch size during training was set to 1. The Adam (Adaptive Moment Estimation) optimizer was used to optimize the situational awareness model parameters. This algorithm dynamically adjusts the learning rate of each parameter by estimating the first moment (mean) and second moment (uncentered variance) of the gradient, thereby improving convergence speed and enhancing the stability of the training process. The initial learning rate was set to 0.001. To further improve the training effect of the situational awareness model, an adaptive learning rate adjustment strategy was adopted: after every 5 training epochs, the learning rate was reduced by 70%. In addition, an early stopping mechanism was introduced to save training time and prevent overfitting. Specifically, if the average loss of the situational awareness model on the validation set did not decrease further within 20 consecutive epochs, the training process was terminated early.

[0113] S4. Construct and train a spatiotemporal forecasting (STF) model. The inputs include historical observed rainfall values, forecasted rainfall values ​​for a single future time step, and historical functional loss values ​​for emergency response areas. The output is the predicted functional loss value for an emergency response area for a single future time step. The prediction results for multiple future time steps are obtained through iterative forecasting.

[0114] In this embodiment, as Figure 4As shown, the spatiotemporal prediction model includes a rainfall feature extraction module, a spatiotemporal feature capture module, and an emergency response area functional loss prediction module. The rainfall feature extraction module, serving as the input module to the spatiotemporal prediction module, processes rainfall data containing multiple historical time steps and one future time step, and concatenates the processed rainfall data with the corresponding historical time step's emergency response area functional loss tensor along the time dimension. The spatiotemporal feature capture module processes time-series data with a graph structure, enabling each graph node to incorporate features from its corresponding spatiotemporal domain. The emergency response area functional loss prediction module outputs the emergency response area functional loss prediction result with the same number of time steps as the input future rainfall time steps.

[0115] Specifically, the rainfall feature extraction module in the spatiotemporal prediction model includes two types of input data: one containing historical data and the other containing other data. A time step and the future F =Rainfall data tensor at 1 time step and corresponding history Functional loss tensor of emergency response area for 24 hours ,in, Integer and , This indicates the batch size used in training the spatiotemporal prediction model. This indicates that the input channel size is 1, corresponding to the feature dimension of functional loss in the emergency response area, and each time step contains... Functional loss information for each emergency response zone node. The rainfall feature extraction module includes an LSTM module and a fully connected layer following the LSTM module, and the rainfall tensor... First, after dimensional rearrangement, it becomes... The input is then fed into an LSTM module, where the input feature dimension is 1, the hidden state dimension is set to 64, and the number of stacked layers is 1. Then, the hidden state of the last time step is extracted from the LSTM output, and after adding a dimension, it becomes... After processing through a fully connected layer, a tensor is obtained. ;Will and By concatenating the data along the time dimension, the final input tensor is formed. .

[0116] Specifically, the spatiotemporal feature capture module includes a two-stage cascaded spatiotemporal convolutional module and an output layer. The two-stage cascaded spatiotemporal convolutional module is used to progressively extract deep spatiotemporal features from the input sequence, and the output layer is used for feature compression and output mapping, and outputs the functional loss prediction result for a future single time step. Each spatiotemporal convolutional module includes two gated temporal convolutional layers and a spatial graph convolutional layer located between the two gated temporal convolutional layers.

[0117] Gated temporal convolutional layers are used to perform convolution operations on time-series signals and combine different activation functions to achieve dynamic feature modeling. They employ one-dimensional causal convolution with a kernel width of [missing value]. Furthermore, nonlinearity is introduced through gated linear units. Specifically, for each node in the graph, the gated-time convolution operation checks the input temporal sequence for nonlinearity during each slide. Having only a few adjacent elements without additional padding will result in a reduction in the length of the time series after each convolution operation. .

[0118] In spatial graph convolution, a Chebyshev polynomial approximation strategy is used for approximate computation:

[0119] (3)

[0120] In formula (3), Represents the convolution kernel. Represents the graph convolution operator. Represents a graphical signal. Let be the normalized graph Laplace matrix, where It is the identity matrix. For degree matrix, This is the adjacency matrix, which here is the average Pearson correlation coefficient matrix obtained in step S24; This is the scaled Laplace matrix. yes The largest eigenvalue, express The Chebyshev polynomial. Represents the convolution kernel parameters. It is the index of the approximate order of the Chebyshev polynomial. It is the approximate order of the Chebyshev polynomial, and also corresponds to the receptive field size of the graph convolution kernel.

[0121] In the first-level spatiotemporal convolution module, the kernel size of the first gated temporal convolutional layer is 3, and the input channels ( ) is 1, output channel ( The kernel size of the first layer is 32, the activation function is GLU; the input and output channels of the spatial graph convolutional layer are 32, and the Chebyshev polynomial approximation order is 3; the kernel size of the second gated temporal convolutional layer is 3, the input channels are 32, the output channels are 64, and the activation function is ReLU.

[0122] In the second-level spatiotemporal convolution module, the first gated temporal convolutional layer has a kernel size of 3, 64 input channels, 128 output channels, and uses the GLU activation function; the spatial graph convolutional layer has 128 input and output channels and uses a Chebyshev polynomial approximation order. The second gated temporal convolutional layer has a kernel size of 3, 128 input channels, 256 output channels, and uses the ReLU activation function.

[0123] The output layer consists of two gated temporal convolutional layers and one 2D convolutional layer. The kernel size of the first gated temporal convolutional layer is [missing information]. The first layer has 256 input channels and 256 output channels, with the GLU activation function. The second layer, a gated temporal convolutional layer, has a kernel size of 1, 256 input channels, 256 output channels, and a Sigmoid activation function. The second layer has 256 input channels, 1 output channel, and a kernel size of 1. The final output tensor size is... .

[0124] Specifically, in the emergency response area functional loss prediction module, the first step is to... After passing through a fully connected layer with both input and output dimensions of 1, and compressing the second dimension, we obtain the functional loss prediction result for the next time step. .

[0125] In this embodiment, as Figure 5 As shown, training the spatiotemporal prediction model specifically includes:

[0126] (1) All rainfall sequences included in the meteorological variables are divided into training set, validation set and test set according to a preset ratio. Specifically, the training and validation sets are first divided into training set and test set at a ratio of 9:1, and then further divided into training set and validation set at a ratio of 7:3.

[0127] (2) For the rainfall sequence included in the training set and the spatial distribution of the functional loss values ​​of all emergency response area nodes at each time step generated by step S2, a sliding time window mechanism is used to construct training sample pairs. The input of each training sample pair includes historical data. Rainfall time-series data at the current time step and one future time step, as well as historical data. The functional loss values ​​of each emergency response area at each time step are output as the predicted functional loss values ​​of each emergency response area at the next time step. Preferably, here... .

[0128] (3) In this embodiment, in order to construct standardized training sample pairs, for each 24-hour rainfall sequence and its corresponding regional functional loss value sequence First, the rainfall sequence is forward-filled: 12 zero-value time steps are introduced before its start, denoted as... And set the functional loss value for all emergency response areas at the corresponding time step. Set the time step to 0 to construct a complete initialization state. When generating training sample pairs, set the time step... to Used as temporal input features in spatiotemporal prediction models to This serves as the spatial feature input. The spatiotemporal prediction model uses this to predict time steps. The distribution of functional losses in each emergency response zone is denoted as... and the simulated values The loss is calculated by comparison. Then, the sliding time window is advanced by one hour, and the above process is repeated until all sample pairs are generated. For each 24-hour rainfall sequence, this strategy will generate 24 sample pairs. The mean squared error function of all generated sample pairs is used as the training loss for the spatiotemporal prediction model, iterating through all rainfall sequences in the training set.

[0129] (4) Validate and test the trained spatiotemporal prediction model, and calculate the validation loss and test loss respectively. For the rainfall sequence included in the validation set and the spatial distribution of the functional loss values ​​of all emergency response area nodes at each time step generated by step S2, calculate the validation loss of the spatiotemporal prediction model in the same way as above; for the rainfall sequence included in the test set and the spatial distribution of the functional loss values ​​of all emergency response area nodes at each time step generated by step S2, calculate the test loss of the spatiotemporal prediction model in the same way as above.

[0130] In this embodiment, the spatiotemporal prediction model is also implemented in the Python environment using the PyTorch 2.0 framework. During training, mean squared error is used as the loss function to measure the deviation between the predicted result and the actual value. The training batch size is set to 64. The model parameter optimization uses the Adam optimizer, which achieves adaptive adjustment of the learning rate of each parameter by simultaneously estimating the first moment (mean) and second moment (uncentered variance) of the gradient, thereby improving the convergence speed and enhancing the stability of the training process. The initial learning rate is set to 0.001 as the starting parameter configuration for model optimization. An adaptive learning rate adjustment strategy is also adopted during the training of the spatiotemporal prediction model, that is, the learning rate is adjusted to 70% of the current value after every 5 training epochs. In addition, an early stopping mechanism is introduced to save model training time, with a tolerance value set to 20, that is, if the average loss of the model on the validation set does not continue to decrease for 20 consecutive training epochs, the training is terminated.

[0131] In step S4, prediction results for multiple future time steps are obtained through iterative forecasting, specifically including:

[0132] When making real-time forecasts of functional losses in emergency response areas at various times corresponding to a 24-hour rainfall sequence, a selection-based forecasting strategy is adopted, and the rainfall sequence is denoted as... ;

[0133] When predicting the functional loss value at the first time step, fill in the gap before the first rainfall value. A zero value, preferably, here , recorded as And supplement the functional loss values ​​of the emergency response areas corresponding to these 12 moments. Both are zero values, obtained by inputting the rainfall sequence. and the corresponding historical time of emergency response area functional loss value get Prediction results corresponding to time steps ;

[0134] When predicting the functional loss value at the second time step, the rainfall sequence used becomes ,Will Prediction results corresponding to time steps As part of the input, namely, the historical moment-in-time emergency response area functional loss value used. ,get Prediction results corresponding to time steps ;

[0135] This process is repeated 24 times until the time-series prediction results of functional loss for all emergency response areas under the entire 24-hour rainfall sequence are generated.

[0136] S5. In practical applications, the real-time disaster monitoring module acquires localized flooding observation information at various times, determines the observed emergency response areas and their observed functional loss values, uses the situational awareness model to infer the functional loss of unobserved emergency response areas, and replaces the predicted values ​​of the corresponding spatiotemporal prediction model with the functional losses of all emergency response areas inferred at different times, outputting the updated multi-time forecast results. Specifically, the steps include:

[0137] S51. Obtain local waterlogging observation information at different times through the real-time disaster monitoring module, and unify the local waterlogging observation information collected within 30 minutes before and after each hour into the local waterlogging observation information corresponding to that hour.

[0138] S52. Based on the local waterlogging observation information extracted by the disaster real-time monitoring module at different hourly times, when more than 70% of the potentially affected buildings in a certain emergency response area are within the coverage of the local waterlogging observation information at a certain time, it is considered as an observed emergency response area, and the proportion of the observed emergency response areas at that time is obtained, which is the observation proportion. Then, the observation function loss value corresponding to all observed emergency response areas at the corresponding time is calculated, and then the prior function loss value of all unobserved emergency response areas is calculated by formula (2), and the function loss value corresponding to all emergency response area nodes is concatenated with their corresponding node-related attribute variables and environmental variables to construct complete node feature information. The situational awareness model trained with the observation ratio closest to that moment is selected. The complete node feature information constructed above is input into the situational awareness model to infer the global emergency response area functional loss at that moment. For emergency response areas that are not observed, the functional loss value output by the situational awareness model is used to predict the result. For emergency response areas that are observed, the observed functional loss value is directly retained to ensure that the output result is consistent with the result obtained based on the observation data.

[0139] S53, Assume the measured rainfall process starts from From the moment on, if At some hour after the time When the proportion of the observed emergency response areas is >5%, the forecast update time is set to [date / time]. Half an hour had passed, because that was exactly the time to complete the interview with the representatives. The collection of localized flooding observation information at any given time; the rainfall input sequence used by the spatiotemporal prediction model at this time is " Time to "Measured hourly rainfall during the specified time period" and " A 24-hour rainfall sequence is constructed by splicing together the predicted rainfall after a certain time. First, the iterative forecasting strategy described in step S4 is used to obtain... Forecast values ​​of functional loss in emergency response areas at each time point after the specified time. Then, obtain through step S52 Real-time functional loss of global emergency response area based on situational awareness model ,use replace Using iterative forecasting strategies to obtain the forecast of The forecast update results for the functional loss of the emergency response area at that time.

[0140] The following example uses a rainstorm event that occurred in a coastal city (let's call it City XX) from September 5th to 6th of a certain year to demonstrate the practical application effect of the dynamic update method for predicting functional loss of building complexes that integrates social media data described in this embodiment.

[0141] Meteorological variables included measured rainfall data from 13 typhoon events within the province to which XX city is located. Rainfall sequences with a 24-hour cumulative rainfall exceeding 350 mm were selected for analysis, resulting in 752 rainfall sequences used for subsequent research. The 24-hour cumulative rainfall of these sequences ranged from 350.9 mm to 890.6 mm. Based on the summarized spatiotemporal evolution of building functional status, 553 potentially affected buildings were identified within the study area. These potentially affected building groups constituted the main research objects for subsequent analysis. Then, using the K-means spatial clustering method, all potentially affected buildings were divided into 50 emergency response zones. The spatial distribution of each emergency response zone is shown below. Figure 6 As shown. Figure 7 The data shows the quantity distribution statistics of the four building types in each emergency response zone. At this time, the average number of buildings in each emergency response zone is about 11.

[0142] Based on the obtained 752 24-hour rainfall sequences, firstly, under the influence of a single rainfall event, the Pearson correlation coefficient matrix between each emergency response area was calculated according to the time history curves of functional loss in each area. Subsequently, the average value of the correlation coefficient matrices obtained for all rainfall events was taken, and this average value was used as a measure of the correlation of functional loss between emergency response areas. The average Pearson correlation coefficient matrix between emergency response areas is shown below. Figure 8 As shown.

[0143] A web crawler program was developed using the Scrapy framework based on the Python language to automate data collection through keywords. For example, each piece of data from a social media platform includes metadata such as user ID, text content, timestamp, and image attachments. In addition to the data obtained through web crawling, to further expand the amount of information that can reflect real-time disaster conditions, this embodiment also obtained relevant data from citizen assistance forms on certain information platforms and various news platforms. During the data collection process, "heavy rain," "torrential rain," and "waterlogging" were consistently used as search keywords.

[0144] Figure 9 The data displays hourly forecasts and measured rainfall, as well as the activity of available social media data, from 16:00 on September 5th to 16:00 on September 6th. Considering that the time resolution of both measured rainfall data and simulated flood scenarios is 1 hour, while the time resolution of social media data is typically 1 minute or 1 second, this embodiment assumes that the social media data collected within 30 minutes before and after each hour corresponds to the flooding scenario at that hour.

[0145] Figure 10 This example demonstrates a typical application scenario of the dynamic update method for predicting functional loss in building clusters that integrates social media data. Prior to the extreme rainstorm event triggered by Typhoon TF1 on September 5-6, the study area (XX City) experienced another heavy rainfall event triggered by a typhoon (denoted as TF2) from July 29-30 of that year. Figure 10 As shown on the left, this event provided the foundation for the construction and calibration of the urban flooding physical model in this embodiment. In the event of July 29th, the urban flooding model constructed within the study area was initially calibrated and validated based on the flooding location information released by the XX City Traffic Police Department. Subsequently, based on the calibrated urban flooding model, the spatiotemporal evolution of functional loss values ​​in various emergency response areas under different rainfall scenarios was simulated. These simulation results are considered to contain the spatiotemporal correlation between functional loss values ​​in different emergency response areas, and this embodiment uses them as the dataset for training the situational awareness model and the spatiotemporal forecasting model.

[0146] like Figure 10 As shown on the right, when the research perspective focuses on the rainstorm event in XX City on September 5th, the dynamic update method for predicting the functional loss of building clusters proposed in this embodiment, which integrates social media data, aims to integrate real-time observation data to minimize the systematic bias introduced by insufficient calibration data in terms of quantity or quality, thereby improving the accuracy of predicting future functional losses in each emergency response area.

[0147] In this embodiment, the forecast update is triggered half an hour after each full hour. Taking the forecast update released at 23:30 on September 5th as an example, by this time, the collection of social media data related to the flooding scenario before and after 23:00 has been completed. Based on this observation data, and using a situational awareness model, the functional loss value of the global emergency response area at 23:00 is inferred and used to replace the functional loss value for 23:00 in the original spatiotemporal prediction model. Subsequently, combining the rainfall input sequence constructed from "measured rainfall up to 23:00" and "forecasted rainfall after 23:00," the spatiotemporal prediction model is used to generate updated forecast results of the functional loss of the emergency response area at future times.

[0148] This study combines Figure 9 The hourly rainfall and the amount of available social media data shown are analyzed. The period from 22:00 on September 5 to 02:00 on September 6 is selected as a typical time window to demonstrate the dynamic updating effect of integrating social media data on the forecast results of functional loss in the area. Figure 11 The study demonstrates the spatial distribution of observed water depth generated by combining waterlogging points extracted from social media data with a localized water depth extension algorithm.

[0149] Figures 12-15 This paper presents the spatiotemporal distribution maps and corresponding statistical histograms of functional loss values ​​for emergency response zones at four time points from 23:00 on September 5th to 02:00 on September 6th. At each time point, the functional loss values ​​for a zone consist of two sources: observed emergency response zones are marked with a thick cyan border, and their functional loss values ​​are the average of the sum of probabilities of all buildings within the observed water depth coverage area being in functional states III to VII; unobserved emergency response zones have their functional loss inferred from a situational awareness model trained at the observation ratio closest to the current observation ratio. The statistical histogram results show that the distribution of functional loss values ​​for each zone at each time point generally shows a trend of gradually shifting towards higher values, reflecting a continuous increase in the overall functional loss level of each zone over time.

[0150] Figure 16 The forecast updates, incorporating social media data at different start times, are displayed, with all forecast results highlighted in blue. To assess the accuracy of the forecasts at different start times, Figure 16The report also displays the function loss results for the observed local areas at three specific times between 00:00 and 02:00 on September 6th, all marked with red borders. The mean and standard deviation of the function loss calculated for each local observed area at different times are listed in the table at the bottom left. The results show that as the reporting time gets closer to the target time, the forecast results for the local observed areas at different times are closer to the observed results, thus verifying the positive role of integrating social media data in improving forecast accuracy.

[0151] The above description only outlines the basic principles and preferred embodiments of the present invention. Those skilled in the art can make many changes and modifications based on the above description, and these changes and modifications should fall within the protection scope of the present invention.

Claims

1. A method for dynamically updating building complex functional loss prediction with fused social media data, the method comprising: The method comprises the following steps: S1, constructing a disaster situation real-time monitoring module, which is used for collecting and processing disaster-related social media data in real time, and extracting local waterlogging observation information in a research area; S2, preparing training data: dividing building types into four categories: high-rise, multi-story, villa and stacked villa, simulating the spatio-temporal evolution of building group function state probability distribution, dividing emergency response areas, and generating functional loss data sets of emergency response areas under various rainfall scenarios; S3, constructing and training a situation awareness model, obtaining the observed functional loss value of the observed emergency response area at the corresponding time according to the local waterlogging observation information, and inferring the functional loss of all unobserved emergency response areas; wherein the situation awareness model comprises two layers of multi-head graph attention layers and ReLU activation functions connected after each layer of multi-head graph attention layers, each layer of multi-head graph attention layer is configured with a plurality of attention heads and multi-dimensional hidden layers, and the output layer of the situation awareness model is a graph attention layer with a dimension of 1; when training the situation awareness model, a random mask strategy is introduced; S4, constructing and training a spatio-temporal prediction model, inputting historical observed rainfall values, future single time step forecast rainfall values and historical emergency response area functional loss values, outputting future single time step emergency response area functional loss prediction values, and obtaining prediction results of multiple time steps in the future through iterative prediction; S5, in actual application, the local waterlogging observation information at each time is obtained through the disaster situation real-time monitoring module, the observed emergency response area and its observed functional loss value are determined, the functional loss of the unobserved emergency response area is inferred by using the situation awareness model, and all emergency response area functional losses observed and inferred at different times are used to replace the prediction values of the spatio-temporal prediction model at the corresponding time, and the updated prediction results of multiple time steps in the future are output.

2. The fused social media data building function loss prediction dynamic updating method of claim 1, wherein, In step S2, the training data includes the training data required by the situation awareness model and the spatio-temporal prediction model, in order to obtain the training data, the required input data includes meteorological variables, flood disaster environment variables and building attribute related variables; wherein the meteorological variables include 24-hour rainfall data measured by the rainfall station, and the rainfall sequences with 24-hour cumulative rainfall exceeding the preset value are selected as the analysis objects; the flood disaster environment variables include DEM data, slope direction data, curvature data, slope data, terrain humidity index data, distance to river center line and distance to pipe network outlet; the building attribute related variables include the outline of each building in the building group, the number of floors, the floor area and the building type.

3. The fused social media data building function loss prediction dynamic updating method of claim 2, wherein, Step S2 specifically comprises: S21, based on the meteorological variables and flood disaster environment variables of the research area, coupling SWMM and WCA2D model to simulate urban waterlogging inundation, and obtaining waterlogging evolution process; S22, using building function state probability prediction method to simulate the spatio-temporal evolution of building group in different function state probability distribution under multiple rainfall events; the building function state includes I, II, III, V and VII, and the larger the value is, the more serious the building is affected. S23, define the building with the lowest probability of being in functional state I at any time under any rainfall sequence as a potential affected building, and divide all potential affected buildings into multiple emergency response zones using the K-means spatial clustering method; define the functional loss of a zone as the average of the sum of the probabilities of all potential affected buildings in the zone being in functional states III-VII, thereby constructing a functional loss dataset of emergency response zones at different times under multiple rainfall scenarios; S24, based on the rainfall sequence, calculate the functional loss time history curve of each emergency response zone under each single rainfall, and calculate the Pearson correlation coefficient matrix between zones based on the time history curve, and use the average Pearson correlation coefficient matrix under all rainfall events as a measure of functional loss correlation.

4. The fused social media data building function loss prediction dynamic updating method of claim 1, wherein, In step S3, training the situational awareness model specifically includes: (1) The input variables of the situational awareness model are , where represents the total number of emergency response patches, 1 represents the dimension of the function loss value, represents the dimension of the node-related attribute, and the node-related attribute includes the proportion of four building types, represents the dimension of the environmental variable, and the environmental variable includes DEM, curvature, distance to the pipe network outlet, and distance to the river center line; the observed emergency response patch is defined as the patch in which more than a certain proportion of potential affected buildings are located within the observed waterlogging coverage range, and the observed function loss value is calculated according to the observed water depth; for each unobserved emergency response patch, the prior function loss value is calculated based on all observed function loss values by the inverse distance weighting method; the output of the situational awareness model is , which represents the predicted function loss value of all emergency response patches at time , and finally only the prediction results for unobserved emergency response patches are retained; (2) For all rainfall sequences contained in the meteorological variables, divide them into training set, validation set and test set according to the preset proportion; from the functional loss value spatial distribution of each time step generated by step S2 for the rainfall sequences in the training set, extract the data in turn, and splice the corresponding node-related attribute variables and environmental variables to construct complete node feature vectors; independently train the situational awareness model under a plurality of preset observation ratios, and calculate the training loss; (3) Verify and test the situational awareness models trained under several preset observation ratios, and calculate the validation loss and test loss respectively.

5. The fused social media data building function loss prediction dynamic updating method of claim 4, wherein, In step (2), when training the situational awareness model, a random mask strategy is introduced to simulate the actual situation that only part of the emergency response zones can be observed, specifically including: (1) For the functional loss value spatial distribution results of all emergency response zone nodes at each time step generated by step S2 for the rainfall sequences contained in the training set, randomly select a number of emergency response zone nodes at a preset observation ratio, retain the simulated functional loss values, and mask the remaining nodes, i.e. mask the functional loss values of the remaining nodes; (2) Calculate the prior functional loss value of all masked nodes according to step (1), which is used to fill the feature vectors of the corresponding masked zone nodes, and splice the functional loss values of all emergency response zone nodes with their corresponding node-related attribute variables and environmental variables to form the input of the situational awareness model; (3) Predict the masked nodes through the situational awareness model, and after the prediction is completed, only the prediction results of the masked zone nodes and the simulation values are calculated by the mean square error loss function.

6. The fused social media data building function loss prediction dynamic updating method of claim 1, wherein, In step S4, the spatio-temporal prediction model comprises a rainfall feature extraction module, a spatio-temporal feature capturing module and an emergency response functional loss prediction module of a functional area; the rainfall feature extraction module is an input module of the spatio-temporal prediction model, used for processing rainfall data containing a plurality of historical time steps and one future time step, and splicing the processed rainfall data with an emergency response functional loss tensor of a corresponding historical time step along a time dimension; the spatio-temporal feature capturing module is used for processing time series data with a graph structure so that each graph node can fuse features from a corresponding spatio-temporal domain; and the emergency response functional loss prediction module is used for outputting an emergency response functional loss prediction result of a time step number consistent with the input future rainfall time step number.

7. The fused social media data building function loss prediction dynamic updating method of claim 6, wherein, The rainfall feature extraction module in the spatio-temporal prediction model includes two types of input data, respectively, a rainfall data tensor containing historical time steps and future F =1 time steps , and an emergency response functional loss tensor of the corresponding historical hours , wherein is an integer and , represents the batch size used in the training of the spatio-temporal prediction model, represents that the input channel size is 1, corresponding to the feature dimension of the emergency response functional loss, and each time step contains the functional loss value information of emergency response functional loss nodes; the rainfall feature extraction module includes an LSTM module and a fully connected layer connected after the LSTM module, the rainfall data tensor is input into the LSTM module and processed by the fully connected layer to obtain a tensor , and are spliced along the time dimension to form the final input tensor .

8. The fused social media data building function loss prediction dynamic updating method of claim 6, wherein, The spatio-temporal feature capturing module comprises two-stage spatio-temporal convolution modules and an output layer in series, the two-stage spatio-temporal convolution modules are used for gradually extracting deep spatio-temporal features in the input sequence, the output layer is used for feature compression and output mapping, and a functional loss prediction result of a future single time step is outputted; each spatio-temporal convolution module comprises two gated temporal convolution layers and a spatial graph convolution layer between the two gated temporal convolution layers, wherein the gated temporal convolution layer is used for convolution operation on a time series graph signal in a time dimension, and realizes dynamic feature modeling in combination with different activation functions, adopts one-dimensional causal convolution, and introduces nonlinearity through a gated linear unit, and the spatial graph convolution adopts a Chebyshev polynomial approximation strategy for approximate calculation; and the output layer comprises two gated convolution layers and a two-dimensional convolution layer.

9. The fused social media data building function loss prediction dynamic updating method of claim 7, wherein, The emergency response subarea function loss prediction module comprises a fully connected layer with an input dimension and an output dimension of 1, and the tensor output by the spatiotemporal feature capturing module is subjected to the fully connected layer with the input dimension and the output dimension of 1 and compression of the second dimension, to obtain a function loss prediction result for a future 1 time step .

10. The fused social media data building function loss prediction dynamic updating method of claim 7, wherein, In step S4, training the spatio-temporal prediction model specifically comprises: (1) dividing all rainfall sequences contained in meteorological variables into a training set, a validation set and a test set according to a preset proportion; (2) For the rainfall sequence contained in the training set and the function loss value space distribution of all emergency response area nodes at each time step generated by step S2, a sliding time window mechanism is used to construct training sample pairs. The input of each training sample pair includes the historical time step and the future 1 time step rainfall time series data, and the historical time step function loss value of each emergency response area, and the output is the function loss prediction value of each emergency response area in the future 1 time step; (3) In Fill in the time step before the rainfall value. A zero value, and supplement The functional loss value of the emergency response area corresponding to each time step is zero, and the spatiotemporal prediction model predicts... The functional loss of each emergency response area at each time step is calculated using the mean square error loss function. Then, the sliding time window is advanced by one time step, and the above process is repeated. For each 24-hour rainfall sequence, 24 training sample pairs are generated. All rainfall sequences in the training set are traversed, and the average mean square error function of all generated sample pairs is used as the training loss of the spatiotemporal prediction model. (4) verifying and testing the trained spatio-temporal prediction model, and calculating a validation loss and a test loss, respectively.

11. The fused social media data building function loss prediction dynamic updating method according to claim 7 or 10, characterized in that, In step S4, the prediction result of a plurality of future time steps is obtained through iterative prediction, specifically comprising: Let the rainfall sequence be denoted as When predicting the function loss value at the first time step, fill in zero values before the first rainfall value, denoted as , and supplement the function loss value of the emergency response area corresponding to this time point, which is also zero, to obtain the prediction result corresponding to the time step by inputting the rainfall sequence and the function loss value of the emergency response area corresponding to the historical time point . When predicting the functional loss value at the second time step, the rainfall sequence used becomes , Will Prediction results corresponding to time steps As part of the input, namely, the historical moment-in-time emergency response area functional loss value used. ,get Prediction results corresponding to time steps ; By analogy, the above steps are repeated for 24 times until the functional loss time series prediction result of all emergency response functional areas under the whole 24-hour rainfall sequence is generated.

Images