Urban road network traffic flood vulnerability assessment method based on multi-source multi-modal data

Abstract

The application discloses a kind of urban road network traffic flood vulnerability evaluation method based on multi-source multi-modal data, through multi-source data fusion technology, combine hydrology, water conservancy, land use, DEM, road network attribute, traffic flow, social economy, social media and mobile phone signaling and other multi-source data, constructs road network disaster vulnerability evaluation model. Weighted fusion method is used to efficiently integrate multi-modal data, and high-performance hydrodynamic integrated model HiPIMS is used to realize high-resolution water depth grid dynamic simulation, accurately capture the changes of road traffic flow, speed and traffic capacity under flood situation, and verify the reliability of its results through verification index. Based on the service characteristics of road grade and flood loss rate, the vulnerability curves of different grades of roads are derived, and the speed is verified by using mobile base station-based mobile phone signaling data, which innovatively realizes the quantitative analysis and grade division of road network disaster vulnerability, significantly improves the accuracy and timeliness of road network risk quantification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban road network traffic flood disaster risk assessment, in particular to a kind of urban road network traffic flood vulnerability assessment method based on multi-source multi-modal big data technology. BACKGROUND

[0002] Current urban flood disaster occurs frequently, and causes serious influence to the traffic capacity and transport function of urban road network. As one of the main disaster bodies of flood disaster, urban road often faces problems such as infrastructure damage, traffic congestion and interruption in flood event, and traffic accidents, economic losses and casualties caused by extreme rainfall are particularly serious. However, the existing flood risk assessment method has significant limitations, mainly in the form of low assessment accuracy and timeliness caused by lack of data or insufficient integration ability.

[0003] Traditional researches mainly rely on meteorological and hydrological data, ignoring comprehensive factors such as impervious surface area and drainage system capacity, resulting in limited accuracy and response efficiency of the results; the existing hydrological and hydrodynamic model has low calculation efficiency and is difficult to simulate complex dynamic flood process. In the road network vulnerability assessment, macroscopic research usually uses the binary judgment method of "unobstructed-interrupted", ignoring the complex topological structure of road network and the change of road dynamic traffic capacity; microscopic research uses fixed water depth threshold to judge road traffic state, but fails to reflect the actual traffic characteristics, for example, some flooded roads still have certain traffic capacity, and wet and slippery road surface that is not flooded may also cause traffic capacity to weaken due to the decrease of friction coefficient. In addition, the existing research generally lacks effective flood verification data, making it difficult to accurately assess the real impact of complex dynamic flood on road network. SUMMARY

[0004] In view of the above technical problems in the related art, the present application provides a kind of urban road network traffic flood vulnerability assessment method based on multi-source multi-modal data, which can solve the above problems.

[0005] To achieve the above technical purpose, the technical scheme of the present application is as follows:

[0006] A kind of urban road network traffic flood vulnerability assessment method based on multi-source multi-modal data, comprising the following steps:

[0007] S100, collect and preprocess multi-source multi-modal traffic disaster data;

[0008] S200, constructing an index system of vulnerability for the pre-processed disaster data, including index extraction and index quantization, wherein the extracted indexes include flood characteristic indexes, vulnerability indexes, and traffic flow data, the index quantization is to extract flood and range characteristics from different types of data by using a convolutional neural network method, and to fuse the characteristics extracted from multi-source and multi-modal data by using a weighted fusion method;

[0009] S300, constructing a road network disaster vulnerability model, analyzing road network disaster vulnerability by using the road network disaster vulnerability model, and classifying road network flood disaster vulnerability;

[0010] S310, dividing roads into different grades according to social and economic function attributes, road network connectivity, and spatial distribution geometric characteristics of the roads;

[0011] S320, fitting an existing water depth-safety speed curve under similar city conditions by using a quadratic function through a curve fitting toolbox in Matlab;

[0012] S330, on the basis of collecting original maximum safe speed, water depth data, road capacity, and traffic flow data of different road grades, using a curve regression fitting method to deduce the specific relationship between safe driving speed and water depth under different road grades;

[0013] S340, using a high-performance hydrodynamic integrated model HiPIMS to simulate and obtain real-time water depth distribution of road network traffic;

[0014] S350, obtaining / estimating road section speed based on mobile phone signaling data;

[0015] S360, inputting road grade, submerged water depth, and traffic flow data into a GIS platform, using a spatial analysis tool to classify road flood disaster vulnerability by using road maximum safe speed and reduction amplitude of flood disaster impact response, and dividing road vulnerability into five grades of low, secondary, medium, high, and very high, and drawing a flood disaster vulnerability grade map of each road section.

[0016] Further, the multi-source and multi-modal traffic disaster data in step S100 includes social media data, traffic data, social and economic data, land use type, DEM, and flood-related road network disaster data obtained in real time through Internet of Things and remote sensing.

[0017] Further, the pre-processing of the collected multi-modal traffic disaster data includes image cleaning and labeling for image data, word segmentation and syntax analysis for text data, and reclassification and grade division for traffic data.

[0018] Further, the flood characteristic index in step S200 is obtained by the HiPIMS model simulation, containing real-time water depth data, and the damaged disaster situation is evaluated by the vehicle speed reduction amplitude calculated by the flood inundation map; the traffic flow data is based on the personnel movement trajectory information in the mobile base station mobile phone signaling data.

[0019] Further, the fusion of the features extracted from the multi-source multi-modal data in step S200 specifically includes: first, evaluating the reliability according to the historical performance, authority, coverage and update frequency of the data source, and evaluating the accuracy of the data through error analysis and time extension, and assigning corresponding weights to each data feature based on the evaluation results; then, the weighted average calculation is performed on the feature values from different data sources in the same index to generate the final fused feature value.

[0020] Further, in step S320, the quadratic function is v0=f(h)=ah 2 +bh+c, wherein v0 represents the maximum safe speed that can be accommodated under the inundation water depth h, a, b, c are function parameters related to the initial maximum speed of the road; the specific relationship in step S330 is represented by the formula v0=f(v i , h), wherein v i represents the original maximum safe speed limit of different levels of roads, and h is the inundation water depth; the calculation formula of the flood vulnerability of the road in step S360 is wherein C r represents the vulnerability of the road network, v max represents the traffic capacity of the vehicle under different road levels, is an adjustment coefficient for considering the influence of different road levels and service levels on vulnerability.

[0021] Further, step S300 further includes: S370, using gradient ascent method to optimize the model parameters.

[0022] Further, it further includes: a social media disaster big data road network disaster vulnerability model verification method based on keyword indexing, through flood disaster and disaster degree information mining and keyword quantification, combining accuracy, recall rate and F-score to verify the accuracy of the model, and a real-time verification and acquisition method based on mobile phone signaling data of road traffic speed, using mobile phone signaling data and road weight to estimate real-time traffic speed, analyze traffic flow conditions, and calibrate the model through actual flood and traffic speed data, improve the accuracy and reliability of the path algorithm and parameters.

[0023] The beneficial effects of this invention are as follows: This application utilizes multi-source data fusion technology, combining hydrological, water conservancy, land use, DEM, road network attributes, traffic flow, socio-economic, social media, and mobile signaling data to construct a high-precision, near-real-time road network disaster vulnerability assessment model. A weighted fusion method coupled with the HiPIMS model is employed to achieve high-resolution dynamic simulation of water depth grids, accurately capturing changes in traffic flow, speed, and capacity of road segments under flood scenarios. The reliability of the results is verified through validation metrics (precision, recall, and F-score). Based on road grade service characteristics and flood loss rates, vulnerability curves for different road grades are derived, and rate verification is performed using mobile signaling data based on mobile base stations, innovatively achieving accurate road segment assessment. This invention significantly improves the accuracy and timeliness of flood risk assessment, providing scientific support for flood risk assessment, traffic emergency response, and disaster prevention and relief, effectively reducing the impact of floods on transportation systems and the socio-economic landscape. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] The present invention will now be described in further detail with reference to the accompanying drawings.

[0026] Fig. 1 This is a flowchart of a method for assessing the flood vulnerability of urban road networks based on multi-source, multi-modal data, as described in an embodiment of the present invention.

[0027] Fig. 2 This is the information composition of social media data in disaster scenarios as described in the embodiments of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0029] like Figs. 1-2 As shown, this invention discloses a method for assessing the flood vulnerability of urban road networks based on multi-source, multi-modal data, comprising the following steps:

[0030] Collect multi-source and multi-modal road network traffic data, which includes social media data, traffic data, social and economic data, land use type, DEM, road network attribute and road construction information data, and also includes flood-related data obtained in real time through the Internet of Things and remote sensing. After vectorization, interpolation and normalization processing, standardized data support is provided for model calculation.

[0031] Build a vulnerability index system, which is built from road network capacity, flood inundation characteristics, dynamic response behavior, etc. It mainly includes index extraction and index quantization. The extracted indexes include flood characteristic indexes, vulnerability indexes and traffic flow data, etc. Index quantization covers feature recognition and fusion analysis.

[0032] Build a road network disaster vulnerability model. The model analyzes the impact of flood disasters on urban road network functions, studies the damage of flood characteristics (water depth) on road network service performance (traffic speed and capacity), extracts dynamic characteristics through a high-performance hydrodynamic integrated model (HiPIMS), uses curve regression to build a safety speed-water depth relationship curve, and analyzes vulnerability indexes based on a GIS platform. Through gradient ascent method iterative optimization of model parameters, the evaluation accuracy is improved, and finally the road network disaster vulnerability analysis and grade division are completed.

[0033] Verify the road network disaster vulnerability model. The model verification is a social media disaster big data road network disaster vulnerability model verification method based on keyword indexing. Through flood disaster and disaster degree information mining and keyword quantification, the model accuracy is verified in combination with precision, recall rate and F-score. At the same time, based on the real-time verification and acquisition method of road traffic speed of mobile phone signaling data, real-time traffic speed is estimated using mobile phone signaling data and road weight, traffic flow conditions are analyzed, and the model is calibrated by actual flood and traffic speed data to improve the accuracy and reliability of path algorithm and parameters.

[0034] Embodiment one:

[0035] The social media data in this application comes from platforms such as Weibo and Twitter, containing images, text, geographic location information and timestamps, which are obtained through crawlers or APIs and stored as structured or unstructured data. Image processing technology combined with labeling tools (Labelme) is used to label and extract feature information of target areas (roads, water accumulation); After key semantic extraction, character cleaning and stop word filtering, sentiment analysis is carried out to quantify disaster, and time and location formats are unified to realize data alignment and efficient analysis.

[0036] This application further obtains vectorized urban road network information, including roads, road segments, and intersections, through an electronic map (OpenStreetMap). Simultaneously, it collects relevant data such as DEM, rainfall, land use type, Manning coefficient, and urban drainage capacity. A high-performance integrated hydrodynamic model (HiPIMS) is used to simulate typical rainfall events. After parameter initialization, the model is run on a GPU accelerator to obtain real-time maximum inundation depth. GIS software is used to segment and process the road network information, extracting road segment distribution information. DEM data is optimized by resolution, rainfall data is interpolated using the Thiessen polygon method, the Manning coefficient is calibrated based on land use, and drainage capacity is determined using a generalized drainage node rate to meet the model input requirements.

[0037] Example 2:

[0038] The indicators extracted in this application include flood characteristic indicators (water depth), vulnerability indicators (vehicle speed reduction), and traffic flow data. The flood characteristic indicators are obtained through HiPIMS model simulation and include real-time water depth data; the vulnerability indicators are the vehicle speed reduction calculated from the flood inundation map; and the traffic flow data is based on personnel movement trajectory information from mobile phone signaling data from mobile base stations.

[0039] This application's index quantification encompasses both feature recognition and fusion analysis. Convolutional Neural Networks (CNNs) are used to extract flood depth and extent features from image data. Specifically, this includes cleaning and segmenting social media images to extract flooded areas and estimate water depth. These image features provide crucial support for quantifying the extent and depth of flood disasters. Fusion analysis integrates and quantifies flood disaster information by assigning weights to features from multiple sources and modalities. The process involves: First, assigning weights to each data feature based on the reliability and accuracy of the data source. Remote sensing images, due to their higher accuracy, receive higher weights than social media images or text descriptions. Then, the feature values ​​of the same index (water depth) from different data sources are weighted and averaged to generate the final fused feature value. To ensure the scientific validity of the fusion, the weights are set through historical data accuracy verification or expert experience. Weighted fusion effectively integrates the advantages of data from different sources, reducing the bias that may arise from a single data source, and providing high-quality input data for subsequent comprehensive assessment and model optimization of flood disasters.

[0040] Example 3:

[0041] The core of the road network vulnerability model construction in this application is to derive the relationship between safe speed and water depth under different road grades through regression analysis, and further calculate the vulnerability of the road network. The following are the specific implementation steps of the model construction method:

[0042] Through the curve fitting toolbox in Matlab, the existing water depth-safety speed curve is fitted by using the following quadratic function, the specific formula is as follows:

[0043] v0=f(h)=ah 2 +bh+c;

[0044] Wherein v0 represents the maximum safety speed that can be accommodated under the submerged water depth h, a, b, c are function parameters related to the initial maximum speed of the road. Further considering the original maximum safety speed v i of different road grades, v0 is a binary function about v i and h, the specific formula is as follows:

[0045] v0=f(v i ,h)=f a (v i ,h)+f b (v i ,h)+f c (v i ,h);

[0046]

[0047] On the basis of collecting the original maximum safety speed v i (including main road, secondary road, branch road, etc.), water depth data, road capacity and traffic flow data of different road grades, the relationship between safety driving speed and water depth under different road grades is derived by using curve regression fitting method, the specific relationship is given by the following formula:

[0048]

[0049] According to the safety driving speed under the submerged condition and the original maximum safety speed of different road grades, the road network vulnerability C r is calculated, the specific formula is as follows:

[0050]

[0051] Wherein is the adjustment coefficient, which is used to consider the influence of different road grades and service level on vulnerability, v max represents the capacity of vehicles under different road grades, the maximum safety speed v i defined by different road grades is directly related to the actual road network function grade, the safety speed v0 is related to the actual submerged water depth, which can be obtained by comparing the water depth and the corresponding safety speed relationship curve;

[0052] According to the functional attributes and importance of the roads, the roads are divided into main roads, secondary roads, branch roads and the like different grades, and the service performance differences of different road grades are considered in the calculation of the vulnerability, the main roads usually bear higher traffic flow, and the flood vulnerability of the main roads is more affected, while the branch roads or secondary roads are less affected, combined with the water depth data, the traffic capacity reduction ratio of each road section is calculated, the difference between the safe driving speed under the flood condition and the original maximum safe speed limit is compared, and the vulnerability is determined;

[0053] The road grade, submerged water depth, traffic flow and the like data are input into the GIS platform, spatialization analysis is carried out, it is identified which roads are more seriously affected and which areas can cause traffic interruption, the flood vulnerability of the roads is classified by the maximum safe speed of the roads and the reduction amplitude of the response affected by the flood, the flood vulnerability of the roads is divided into five grades of low grade, secondary grade, middle grade, high grade and very high grade, the flood vulnerability grade map of each road section is drawn, the vulnerability grade of each road section is clear, the decision basis is provided for the urban traffic management and emergency response, and the road network disaster vulnerability model is preliminarily established;

[0054] Further, the gradient ascent method is used to optimize the model parameters. First, the maximum eigenvalue and the corresponding eigenvector of the pair-wise comparison matrix passing the consistency test are calculated by the analytic hierarchy process as the initial value of the model optimization. Subsequently, the gradient ascent method is used for iterative optimization, and the weight combination is adjusted to gradually approach the maximum value of the model accuracy index (area under the ROC curve AUC). The main steps of the gradient ascent algorithm are as follows:

[0055] Preliminary weight setting: the initial weight is set according to the historical data and expert experience;

[0056] Difference calculation: the difference (error) between the model prediction result under the current weight and the true value is calculated;

[0057] Gradient calculation and update: the gradient of the model error with respect to each parameter is calculated, and the weight parameter is updated by the gradient ascent method, and the specific formula is as follows:

[0058]

[0059] Wherein, θ is the model parameter, α is the learning rate, and J(θ) is the loss function;

[0060] Iterative optimization: the model accuracy is gradually approached to the maximum value by adjusting the parameters through iterative optimization.

[0061] In summary, when evaluating the vulnerability of road network using the road network disaster vulnerability model, the reduction amplitude of road safety speed is first taken as the core evaluation index, and the corresponding evaluation model is established. Subsequently, the natural breakpoint method based on ArcGIS is used for grade division. This method calculates the variance and judges the classification effect to ensure that the difference between categories is significant and the difference within categories is minimal. In the application process of the model, the submerged water depth needs to be extracted based on the flood simulation model, and the reduction amplitude of road safety speed under different water depth conditions needs to be quantified, based on which the grade division is carried out, so as to realize the precise evaluation of the vulnerability of road network disaster.

[0062] Example Four

[0063] The verification principle of the road network vulnerability evaluation model in this application is to obtain the actual disaster information (relationship between speed and water depth) by using big data mining technology, to compare the predicted vulnerability level of road network with the actual disaster grade, and to evaluate the accuracy of the model by using the indicators such as precision (Precision), recall (Recall) and F-score.

[0064] Precision measures the accuracy of the model prediction, and represents the proportion of true positive examples in the predicted positive examples in the classification results. If the model evaluates u sub-samples in v samples, and w of them are correct results, then the precision formula is:

[0065] Precision = u / w;

[0066] In the evaluation of road network vulnerability, the positive example refers to the case where the vulnerability level predicted by the model is consistent with the actual disaster grade. Recall measures the recognition ability of the model to positive examples, and represents the proportion of correctly identified positive examples to all actual positive examples, and the specific formula is as follows:

[0067] recall = u / v;

[0068] F-score is the harmonic mean of precision and recall, which considers both precision and recall, and can more comprehensively evaluate the performance of the model. By calculating the precision, recall and F-score values of the model under different rainfall scenarios, the calculation method is as follows:

[0069] F-score = 2 Precision*recall = (Precision+recall).

[0070] Further, road traffic flow (link speed) is an important indicator of urban traffic conditions. Using mobile phone signaling data, user information and signal propagation data are collected through the mobile communication system to estimate the average speed of the link. The model is verified and calibrated using real flood disaster data and traffic flow data to ensure its accuracy and reliability. To obtain real-time monitoring of road traffic speed, the following methods can be used:

[0071] Point matching: Based on map data or GIS information, a database containing all link locations and attributes is constructed. For offset points, the nearest neighbor algorithm is used to calculate the Euclidean distance between each link and the offset point, and the link with the shortest distance is selected as the matching result. This allows the offset point to be accurately mapped to the corresponding link, achieving efficient correction and accurate matching of points.

[0072] Trajectory data generation: After point matching, the points have been accurately mapped to the corresponding links. Combined with the timestamp and latitude and longitude data of the points, the dynamic drawing and path estimation of the moving route of the person along the electronic map navigation path (Gaode API) are performed, generating complete user trajectory information and intuitively displaying the dynamic behavior pattern.

[0073] Speed data generation: Process and analyze the trajectory data to extract the average speed of the link. First, calculate the total distance of each pedestrian moving on the link based on their trajectory data, and determine the total time they spend on the link through the timestamp to calculate the average speed of each pedestrian, the formula is: average speed = total distance / total time. When multiple pedestrians move on the same link, the average speed of all pedestrians within a certain period of time can be aggregated. Further, to more accurately reflect the overall speed distribution of the link, consider the contribution of each pedestrian's travel time on the link to the average speed, and use the weighted average method to calculate the average speed of the link, the formula is: link average speed = ∑(individual speed x travel time) ÷ ∑(travel time).

[0074] Verify the applicability and effectiveness of the model using a specific city as an example. For example, taking Zhengzhou City as the research object, simulate different rainfall intensity flood scenarios to verify the consistency of the model's predicted vulnerability distribution map with the actual disaster situation. Analyze the sources of error and optimize the model performance by adjusting the index weight, parameter and data precision to improve the accuracy and practicality of the evaluation.

[0075] In summary, the application comprehensively considers the influence of various factors on the vulnerability of the road network, including road network structure, function, terrain, rainfall, drainage, etc., and the evaluation results are more accurate and comprehensive. Advanced technical means are adopted, such as GIS technology for data processing and analysis, HiPIMS model for high-precision flood simulation, and big data mining technology for model verification, which improves the scientificity and reliability of the model. The model verification method based on keyword indexing big data mining technology can better reflect the accuracy of the model in actual application, avoiding the problem of unreliable model caused by lack of effective verification data in traditional verification methods. The application can be applied to urban planning and management departments, providing scientific basis for urban flood control and drainage planning, road facility construction and maintenance, and traffic management, etc. In terms of disaster warning and emergency response, the model can quickly evaluate the vulnerability of the road network, providing support for developing reasonable evacuation routes and rescue plans, which helps to reduce the impact of rainstorm waterlogging disasters on urban traffic and residents' life.

[0076] The above merely describes preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for urban road network traffic flood vulnerability assessment based on multi-source multi-modal data, characterized in that, The method comprises the following steps: S100, collecting and preprocessing multi-source multi-modal traffic disaster data; S200, constructing an index system of vulnerability for the preprocessed disaster data, including index extraction and index quantization, wherein the extracted indexes include flood characteristic indexes, vulnerability indexes and traffic flow data, index quantization is to extract floodwater depth and range features from different types of data using a convolutional neural network method, and to fuse the features extracted from multi-source multi-modal data through a weighted fusion method; S300, constructing a road network disaster vulnerability model, analyzing the road network disaster vulnerability through the road network disaster vulnerability model, and classifying the road network flood disaster vulnerability; S310, dividing roads into different grades according to the social and economic function attributes, road network connectivity and spatial distribution geometric characteristics of the roads; S320, using a quadratic function to fit the existing water depth-safety speed curve under similar city conditions through the curve fitting toolbox in Matlab; The quadratic function in step S320 is wherein represents the maximum safe speed that can be accommodated under the submerged water depth h, a, b, c are function parameters related to the initial maximum speed of the road; S330, on the basis of collecting original maximum safe speed, water depth data, road capacity and traffic flow data of different road grades, using curve regression fitting method to deduce the specific relationship between safe driving speed and water depth under different road grades; The specific relationship in step S330 is represented by the formula wherein represents the original maximum safe speed limit of different levels of roads, h is the inundation water depth, , , , , , , , , are function parameters obtained by using curve regression fitting based on the original maximum safe speed limit of different levels of roads, water depth data, road traffic capacity and traffic flow data. S340, using a high-performance hydrodynamic integrated model HiPIMS to simulate and obtain real-time water depth distribution of road network traffic; S350, obtaining / estimating road section speed based on mobile phone signaling data; S360, inputting road grade, submerged water depth and traffic flow data into a GIS platform, using spatial analysis tools, classifying the road flood disaster vulnerability through the maximum safe speed of the road and the reduction rate of the response affected by the flood disaster, and dividing the road vulnerability into five grades of low, secondary, medium, high and very high, and drawing a flood disaster vulnerability grade map of each road section. The formula for calculating the flood vulnerability of the road in step S360 is wherein represents the vulnerability of the road network, represents the traffic capacity of the vehicle under different road grades, is an adjustment coefficient, which is used to consider the influence of different road grades and service levels on the vulnerability.

2. The urban road network traffic flood vulnerability assessment method based on multi-source multi-modal data according to claim 1, characterized in that, The multi-source multi-modal traffic disaster data in step S100 includes social media data, traffic data, social and economic data, land use type, DEM, and flood-related road network disaster data obtained in real time through the Internet of Things and remote sensing.

3. The method of claim 2, wherein, The preprocessing of the collected multi-modal traffic disaster data includes image cleaning and labeling for image data, word segmentation and syntax analysis for text data, and reclassification and grading for traffic data.

4. The method of claim 1, wherein, The flood feature indexes in step S200 are obtained by the HiPIMS model simulation, including real-time water depth data, and the damaged disaster is evaluated by the reduction rate of vehicle speed caused by flood submergence; the traffic flow data is based on the mobile trajectory information of personnel in the mobile phone signaling data of mobile base stations.

5. The method of claim 1, wherein, The fusion of the features extracted from multi-source multi-modal data in step S200 specifically includes: first, evaluating the reliability according to the historical performance, authority, coverage and update frequency of the data sources, and evaluating the accuracy of the data through error analysis and time extension, and assigning appropriate weights to each data feature based on the evaluation results; then, the weighted average calculation is performed on the feature values from different data sources in the same index to generate the final fused feature values.

6. The method of claim 1, wherein, The step S300 further comprises: S370, optimizing the model parameters by using a gradient ascent method.

7. The method of claim 1, wherein, The method further comprises: a social media disaster big data road network disaster vulnerability model verification method based on a keyword index, which verifies the model accuracy through flood disaster and disaster degree information mining and keyword quantification, combined with precision, recall rate and F-score, and simultaneously verifies and obtains the real-time traffic speed based on a mobile phone signaling data road traffic speed estimation method, analyzes the traffic flow condition, and calibrates the model through actual flood and traffic speed data, thereby improving the accuracy and reliability of the path algorithm and parameters.

Images