Urban disaster risk research and judgment method and device

Abstract

The invention discloses an urban disaster risk studying and judging method and device, and the method comprises the steps: obtaining a weather early-warning region according to weather early-warning information, obtaining a flood early-warning region according to water regimen early-warning, and carrying out the spatial superposition of the weather early-warning region and the flood early-warning region, and determining a to-be-studied and judged urban region; the method comprises the following steps: constructing an urban disaster risk keyword library, distinguishing flood, geological disasters and secondary disaster risks caused by rainstorm, constructing a big language model risk research and judgment cue word, obtaining information related to the rainstorm in social media, extracting disaster risk information from the information, and obtaining a potential expansion risk semantic index according to the disaster risk information; abnormal information in urban mobile phone signaling and floating car GPS traffic speed data before and after rainstorm is obtained and quantified, and a potential expansion risk human activity index is obtained according to the abnormal degree; and according to a causal association method, constructing an urban disaster chain semantic association network, fusing multi-source data to update a disaster evolution situation in real time, outputting disaster information and predicting possibly derived secondary disasters.

Description

Technical Field

[0001] This application relates to a method and apparatus for assessing urban disaster risks. Background Technology

[0002] Disaster risk is a scenario of adverse future events primarily caused by natural events or forces. Due to the complexity and dynamism of urban systems, disaster risks are constantly spreading and changing, easily causing losses. In recent years, global climate change and frequent rainstorms have made floods one of the most common urban disasters. Currently, the tools and methods for assessing rainstorm disaster risks lack real-time and dynamic capabilities, still relying on rainstorm warnings issued by meteorological departments and on-site investigations and reports, making it difficult to achieve early and rapid identification of risks and hazards and accurate prediction of the evolution trend of disasters.

[0003] The means of urban disaster risk perception are becoming increasingly sophisticated, providing a multi-dimensional perspective for risk identification. In particular, various types of social perception data have shown great potential: for example, images, texts, and videos spontaneously posted by users on social media platforms have become an important channel for capturing early risks (such as localized water accumulation, fallen trees, and damaged facilities); mobile phone signaling data can reveal dynamic changes in population distribution, abnormal gathering and movement patterns, and commuting disruptions during rainstorms; and traffic big data such as floating car GPS can indirectly reflect traffic anomalies caused by road flooding.

[0004] Despite the increasingly abundant data foundation, existing research still faces the following key challenges: 1. There is a lack of research on early-stage disaster risk assessment, making early-stage risk monitoring and quantification difficult; 2. No model has been established to study the evolution of urban disaster risks under the influence of rainstorms; 3. Often only a single type of data is used, lacking research on disaster risk assessment that integrates multi-source data, and failing to adequately reflect risk information at each stage before, during, and after a disaster, thus making it impossible to comprehensively assess urban disaster risks. Summary of the Invention

[0005] The present invention aims to provide a method and apparatus for assessing urban disaster risks, in order to overcome the shortcomings of the existing technology. The technical problem to be solved by the present invention is achieved through the following technical solution.

[0006] A method for assessing urban disaster risks includes the following steps:

[0007] Meteorological warning areas are obtained based on meteorological warning information, and flood warning areas are obtained based on water situation warnings. The meteorological warning areas and flood warning areas are spatially overlaid to determine the urban areas to be assessed. High, medium and low risk assessment areas are generated and vector boundary data with geographic coordinates are output. The urban areas to be assessed are updated regularly.

[0008] Construct a keyword database for urban disaster risks, differentiate between floods, geological disasters and their secondary disaster risks caused by rainstorms, construct risk assessment prompts using a large language model, obtain information about rainstorms from social media and extract disaster risk information from it, and obtain a potential extended risk semantic index based on the disaster risk information;

[0009] Obtain and quantify abnormal information from urban mobile phone signaling and floating car GPS traffic speed data before and after heavy rain, and obtain a potential extended risk human activity index based on the degree of abnormality.

[0010] Based on the causal association method, a semantic association network of urban disaster chains is constructed, which integrates multi-source data to update the disaster evolution trend in real time and outputs disaster information and predictions of possible secondary disasters.

[0011] Preferably, the hydrological information includes rainfall data obtained from meteorological departments, and upstream and downstream water levels of rivers, lakes and reservoirs, and their early warning information obtained from water conservancy departments.

[0012] Preferably, the step of obtaining information about rainstorms from social media and extracting disaster risk information from there includes:

[0013] Extract risk keywords from information about rainstorms on social media to form a set R = {w1, w2, ..., w n Different risk keywords correspond to different risk levels, and the Potential Extended Risk Index (PRL) is defined as follows:

[0014]

[0015] The risk level of the keyword with the highest risk level is used as the overall potential expansion risk index of the set, and the maximum value of the potential expansion risk index within the grid is used as the potential expansion risk level of that grid.

[0016] Preferably, abnormal information in urban mobile phone signaling and floating car GPS traffic speed data before and after heavy rain is acquired and quantified. Based on the degree of abnormality, a potential extended risk human activity index is obtained, including:

[0017] Collect mobile phone signaling data and traffic speed data from the same location, and collaboratively calculate joint confidence.

[0018] P l =P x *(1+α*P j )

[0019] Among them, P l P represents the joint confidence level of a grid or road, used to indicate the likelihood of potential expansion risk at that location. x P represents the probability of abnormal mobile signaling data. jLet α represent the probability of anomalies in vehicle speed data, and α be the population anomaly weighting factor.

[0020] Preferably, the information reflected by abnormal mobile phone signaling and traffic abnormalities is converted into text descriptions, and a correlation is established with information about rainstorms on social media based on their location and time to form a merged text.

[0021] A pre-trained large language model is used to extract causal relationships from merged text.

[0022] Construct a three-part knowledge framework for urban disaster risk propagation, comprising risk type, transmission relationship, and propagation probability. The propagation probability of each node includes the causative factor X. 致灾 and vulnerability parameter X 易损 The risk type is the potential expansion risk level, and the propagation probability is derived from the propagation probability of historical cases in the grid of this area, mined using a logistic regression model, and calculated by the following formula:

[0023]

[0024] Where β0 is a constant term, β i ε k is the regression coefficient.

[0025] This invention also relates to an urban disaster risk assessment device, comprising:

[0026] The early warning analysis unit obtains the meteorological early warning area based on meteorological early warning information and the flood early warning area based on water situation early warning. It then spatially overlays the meteorological early warning area and the flood early warning area to determine the urban area to be analyzed.

[0027] The multimodal semantic information processing unit constructs a keyword database for urban disaster risks, distinguishes between floods, geological disasters and their secondary disaster risks caused by rainstorms, constructs risk assessment prompts based on a large language model, obtains information about rainstorms from social media and extracts disaster risk information from it, and obtains a potential extended risk semantic index based on the disaster risk information.

[0028] The urban dynamic anomaly detection unit acquires and quantifies abnormal information from urban mobile phone signaling and floating car GPS traffic speed data before and after rainstorms, and obtains a potential extended risk human activity index based on the degree of anomaly.

[0029] The disaster chain simulation unit constructs a semantic association network of urban disaster chains based on causal correlation methods, integrates multi-source data to update the disaster evolution trend in real time, and outputs disaster risk information and predictions of possible secondary disasters.

[0030] The urban disaster risk assessment method of this invention effectively utilizes the advantages of multi-source spatiotemporal big data, such as social media information, traffic big data, and mobile phone signaling, which are characterized by strong real-time performance and high spatiotemporal spatiotemporal characteristics. Through multi-source data fusion and artificial intelligence technology, it establishes a quantitative relationship between public perception information, abnormal human activity data, and disaster risk, and constructs a dynamically updated urban disaster chain projection model, thereby significantly improving the real-time assessment capability of urban disaster risk. It constructs a full-chain risk prediction system covering "pre-disaster - during-disaster - post-disaster." In the pre-disaster stage, potential risks and hazards can be predicted based on meteorological warnings; in the during-disaster stage, accurate disaster assessment is achieved through real-time monitoring of public perception information and urban operational anomalies; and in the post-disaster stage, secondary disaster risks are predicted based on the dynamically updated disaster chain projection model. This technology breaks through the spatiotemporal limitations of traditional disaster assessment, providing intelligent decision support for urban disaster prevention and mitigation, emergency response, and post-disaster reconstruction throughout the entire process and from multiple dimensions. It significantly improves urban safety levels and provides precise spatiotemporal decision support for urban emergency response, possessing significant application value in disaster prevention and mitigation, emergency resource allocation, and other fields. Detailed Implementation

[0031] The technical solution of the present invention will be described in further detail below.

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0033] The detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0034] In the description of this invention, it should be noted that the terms "upper", "lower", "inner", "outer", etc., refer to the orientation or positional relationship of the product when it is used. They are only used to facilitate the description of this invention and to simplify the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0035] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0036] The embodiments of the present invention will be described in detail below.

[0037] A method for assessing urban disaster risks includes the following steps:

[0038] Meteorological warning areas are obtained based on meteorological warning information, and flood warning areas are obtained based on hydrological warning information. The hydrological information includes rainfall data, upstream and downstream water levels of rivers, and water levels of lakes and reservoirs obtained from meteorological departments. The water levels of rivers and reservoirs and their warning information data are obtained from water resources departments. The meteorological warning areas and flood warning areas are spatially overlaid to determine the urban areas to be assessed, generating high, medium, and low risk assessment areas and outputting vector boundary data with geographic coordinates.

[0039] The system receives real-time data on red / orange rainstorm warnings issued by meteorological departments, and simultaneously acquires monitoring data from hydrological stations. When water levels at upstream stations exceed warning levels, it identifies the 1-2 nearest downstream cities as the areas requiring assessment. It spatially overlays meteorological and flood warning areas to automatically generate high, medium, and low-risk assessment areas and outputs vector boundary data with geographic coordinates. A dynamic update mechanism is also established, automatically updating the assessment area every 15 minutes, and immediately triggering recalculation when a new warning is issued or hydrological data changes abruptly.

[0040] A keyword database for urban disaster risks was constructed to distinguish the risks of floods, geological disasters and their secondary disasters caused by rainstorms. Risk assessment prompts were constructed using a large language model. Information about rainstorms from social media was obtained and disaster risk information was extracted from it. Social media data mainly included platforms such as Sina Weibo, Douyin and Kuaishou. Potential extended risk semantic indexes were obtained based on disaster risk information.

[0041] Table 1 shows the keyword database and classification standards for urban safety risks.

[0042]

[0043]

[0044]

[0045] Table 1

[0046] Since this embodiment primarily focuses on risks that have not yet occurred, events at the tail end of the disaster chain can be assigned a lower risk level. However, the possibility of further risk escalation still needs to be considered, and therefore are defined as Level 1 risk. If the following disasters are not involved or the described information is a historical disaster, the risk level is no risk.

[0047] Based on keywords, obtain social media text data containing the keyword "rainstorm" within a certain period of time, and perform deduplication on the data to remove completely identical social media text information.

[0048] Stop word removal: Use a stop word list, including all English letters and common punctuation marks, and use string matching methods to remove stop words from the text.

[0049] Set prompts for a large language model to obtain disaster risk information. The prompts should progressively differentiate between text and image data and include case studies to guide the large language model in outputting standardized answers. For text information, the prompts could be:

[0050] Please extract the names of disaster risk areas and risk information from the text, and output standardized JSON according to the following rules:

[0051] Extract the geographical location of disaster risks according to the following rules: 1. Extract addresses in five levels: province → city → district / county → street → specific location (in this order); 2. Different risk areas need to be separated: addresses with the same name (e.g., complete path XX province XX city XX district XX street XX location) are directly merged into a continuous string; those pointing to different areas are separated by a vertical bar |. Example: Input: Flooding is occurring everywhere in Shijiazhuang; many streets in Chang'an District and Xinhua District are flooded. Output: Hebei Province, Shijiazhuang City, Chang'an District | Hebei Province, Shijiazhuang City, Xinhua District.

[0052] Extract disaster risk information from Table 1, separating multiple disaster risks with |. Example: Input: The river next door has overflowed, flooding is everywhere, and houses will collapse if this continues. Output: type: flood disaster, description: Flash floods or river / lake floods have occurred, potentially leading to secondary disasters such as house collapses. If the above risks are not included, output "None".

[0053] Combining the two steps, the output text information result is as follows: Input: The landslide in Qujing City, Yunnan Province is very serious. Trees this thick have fallen. It feels like a landslide is about to occur! Output JSON is as follows:

[0054]

[0055] Continue utilizing the large language model to assess the risks indicated by images and videos. This requires extracting keyframes from the video every 5 seconds, converting them into images, and then inputting these images into the large language model. For image information, the prompts could be...

[0056] Based on Table 1, identify which risk types and keywords the scenes and text in the images match. Produce a standardized JSON output listing each risk separately, with each item containing only one primary disaster type (e.g., "flood") and a description of its impact. Extraction should be based solely on explicit information from the images, without adding subjective judgment. For example...

[0057]

[0058]

[0059] A potential extended risk semantic index is defined to extract semantic information from social media messages and quantitatively assess the potential for disaster risk expansion. By constructing a risk keyword weight table and employing the maximum weight method or a weighted scoring mechanism, the severity of future losses caused by disasters mentioned in the text can be effectively assessed.

[0060] Table 2 defines the weights of the semantic index for potential extension risks.

[0061]

[0062] Table 2

[0063] Extract risk keywords from information about rainstorms on social media to form a set R = {w1, w2, ..., w n Different risk keywords correspond to different risk levels, and the Potential Extended Risk Index (PRL) is defined as follows:

[0064]

[0065] The keyword with the highest risk level is used as the overall potential expansion risk index of the set, and the maximum value of the potential expansion risk index within the grid is used as the potential expansion risk level of that grid.

[0066] Abnormal information in urban mobile phone signaling and floating car GPS traffic speed data before and after rainstorms is obtained and quantified. Based on the degree of abnormality, a potential extended risk human activity index is obtained.

[0067] Collect mobile phone signaling data and traffic speed data from the same location, and collaboratively calculate joint confidence.

[0068] P l =P x *(1+α*P j )

[0069] Among them, Pl P represents the joint confidence level of a grid or road network, used to indicate the likelihood of potential expansion risk at a certain location. x P represents the probability of abnormal mobile signaling data. j Let α represent the probability of anomalies in vehicle speed data, and α be the population anomaly weighting factor.

[0070] The information reflected by abnormal mobile phone signaling and traffic abnormalities is converted into text descriptions, and the correlation between their location and time and information about rainstorms on social media is established to form merged text information.

[0071] A pre-trained large language model is used to extract causal relationships from merged text information;

[0072] Construct a three-part knowledge framework for urban disaster risk propagation, comprising risk type, transmission relationship, and propagation probability. The propagation probability of each node includes the causative factor X. 致灾 and vulnerability parameter X 易损 The risk type is the potential expansion risk level, and the propagation probability is derived from the propagation probability of historical cases in the grid of this area, mined using a logistic regression model, and calculated by the following formula:

[0073]

[0074] Where β0 is a constant term, β i ε k is the regression coefficient.

[0075] The following explains in detail how to obtain the potential expansion risk human activity index based on the degree of anomaly.

[0076] Using a 7-day period, urban mobile phone signaling and traffic speed data were acquired for 21 days before and after the rainstorm. Mobile phone signaling data was processed in 500m grid units, and traffic speed data was processed in road units, and then time-series data was processed. The mobile phone signaling data was provided by the communication provider with hourly 500m*500m grid population counts, and the traffic speed data was provided by map navigation platforms such as Gaode Maps, with average speeds calculated hourly.

[0077] The Prophet algorithm is used to calculate anomalies in mobile phone signaling or traffic speed data. This algorithm can capture periodic fluctuations and handle the short-term impact of specific dates or time periods, such as holidays, on time series. The formula is as follows:

[0078] y(t,s)=g(t)+s(t)+h(t)+ε(t)

[0079] The trend term g(t) describes the long-term trend of the time series and is typically fitted using a piecewise linear model with an autoregressive and seasonal term. The seasonal term s(t) describes the periodic changes in the time series and is typically fitted using a Fourier series. The holiday term h(t) describes anomalous events in the time series (such as holidays) and is typically represented by a custom binary variable.

[0080] The trend, seasonal, and holiday terms are added together to obtain the predicted value of the time series. ε(t) is the error term, which is usually assumed to be normally distributed.

[0081] The residuals of the output data are calculated using the Prophet algorithm, and the standard deviation of the residuals for that period is calculated using historical data. If the residuals exceed 3σ, proceed to the next step.

[0082] r t =y 真实 -y 预测

[0083]

[0084] r t For mobile phone signaling or traffic speed data residuals, y is the value at a certain point, σ is the standard deviation of the residuals, and N is the total number of data at that moment in the acquired historical data.

[0085] Based on the above formula, the probability of anomalies in mobile phone signaling data or traffic speed data is obtained.

[0086]

[0087] Based on the above formula, the probability of anomalies in mobile phone signaling data or traffic speed data is obtained.

[0088]

[0089] Since the number of people reflected by mobile phone signaling is highly correlated with traffic congestion, mobile phone signaling data and traffic speed data from the same location are taken together to calculate the joint confidence level, thereby improving the ability to identify anomalies.

[0090] P l =P x *(1+α*P j )

[0091] Where P l For the joint confidence of a certain grid or a certain road, P x P represents the probability of abnormal mobile signaling data. j P represents the probability of anomalies in vehicle speed data. α is the population anomaly weighting factor; when focusing on monitoring population anomaly risks, α can be set to 0.3; when focusing on monitoring traffic anomaly risks, P0.3...x and P j By swapping the positions, α is still set to 0.3. This ensures that the probability of anomaly increases when both types of data are abnormal simultaneously, while the value remains essentially unchanged when only a single data point is abnormal.

[0092] A higher confidence level indicates a greater likelihood of potential expansion risk at that location, i.e., according to P... l The potential expansion risk human activity index was obtained, 0. <P l ≤0.3 indicates low risk, 0.3 <P l ≤0.6 is considered medium risk, 0.6 <P l This is considered high-risk. Based on the anomaly directions extracted by the Prophet algorithm, abnormal population concentration / decrease, traffic congestion / disruption, etc., are identified.

[0093] Based on the causal association method, a semantic association network of urban disaster chains is constructed, which integrates multi-source data to update the disaster evolution trend in real time and outputs disaster risk information and predictions of possible secondary disasters.

[0094] Specifically, the following steps are used to achieve this:

[0095] By integrating heterogeneous data through unified text representation, information reflected by abnormal mobile phone signaling and traffic anomalies is transformed into text descriptions, such as abnormal crowd gatherings and mobile phone signal interruptions. Correlation is established with social media text data based on their location, time, and other factors.

[0096] A pre-trained large language model is used to extract causal relationships. The prompts are as follows: Please output disaster risk information with causal relationships. For example: "The road is flooded, causing traffic disruption and people are trapped in the car". The output will be: "Urban flooding - traffic disruption - people trapped".

[0097] Based on domain knowledge, a tripartite knowledge framework for urban disaster risk propagation is constructed, comprising {risk type (potential extended risk level) - transmission relationship - propagation probability}, where the probability will be calculated in subsequent steps. The propagation conditions for each node include {existence factor (existence factor parameter 1, parameter 2, ...), node vulnerability (vulnerability parameter 1, parameter 2, ...)}.

[0098] Using geographic grids as units and UTC time as the reference, the spatiotemporal references of social media risk, traffic risk, and mobile signaling risk information are aligned and unified to WGS84 coordinates.

[0099] The propagation probability of historical cases in the grid of this region was mined using a logistic regression model.

[0100] By using real-time data, we can match the propagation events in the rule graph, activate the corresponding propagation paths, and output the possible risk propagation situation.

[0101] This invention also relates to an urban disaster risk assessment device, comprising:

[0102] The early warning analysis unit obtains the meteorological early warning area based on meteorological early warning information and the flood early warning area based on water situation early warning. It then spatially overlays the meteorological early warning area and the flood early warning area to determine the urban area to be analyzed.

[0103] The multimodal semantic information processing unit constructs a keyword database for urban disaster risks, distinguishes between floods, geological disasters and their secondary disaster risks caused by rainstorms, constructs risk assessment prompts based on a large language model, obtains information about rainstorms from social media and extracts disaster risk information from it, and obtains a potential extended risk semantic index based on the disaster risk information.

[0104] The urban dynamic anomaly detection unit acquires and quantifies abnormal information from urban mobile phone signaling and floating car GPS traffic speed data before and after rainstorms, and obtains a potential extended risk human activity index based on the degree of anomaly.

[0105] The disaster chain simulation unit constructs a semantic association network of urban disaster chains based on causal correlation methods, integrates multi-source data to update the disaster evolution trend in real time, and outputs disaster risk information and predictions of possible secondary disasters.

[0106] It should be noted that the above detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0107] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments described in this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0108] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that the embodiments of this application described herein can be implemented in orders other than those described herein.

[0109] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or apparatus.

[0110] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., may be used herein to describe the spatial positional relationship of a feature to other devices or features. It should be understood that spatial relative terms are intended to encompass not only the different orientations of a device in operation. For example, if a device is inverted, a device described as "above" or "on top of" other devices or structures will subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways, such as rotated 90 degrees or in other orientations, and the spatial relative descriptions used herein will be interpreted accordingly.

[0111] In the detailed description above, similar symbols typically identify similar parts unless the context otherwise indicates otherwise. The embodiments described in the detailed specification and claims are not intended to be limiting. Other embodiments may be used and other changes may be made without departing from the spirit or scope of the subject matter presented herein.

[0112] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for analyzing urban disaster risk, characterized in that: The method comprises the following steps: According to the meteorological warning information, the meteorological warning area is obtained, the flood warning area is obtained according to the water regime warning, the meteorological warning area and the flood warning area are spatially superimposed to determine the urban area required for research and judgment, the high, medium and low risk research and judgment areas are generated, the vector boundary data with geographical coordinates are output, and the urban area required for research and judgment is regularly updated; A city disaster risk keyword library is constructed, rainstorm-induced flood, geological disaster and secondary disaster risk are distinguished, a large language model risk research and judgment prompt word is constructed, information about rainstorm in social media is obtained and disaster risk information is extracted therefrom, and a potential extension risk semantic index is obtained according to the disaster risk information; Abnormal information in the traffic speed data of the city mobile signaling and the floating car GPS before and after the rainstorm is obtained and quantified, and a potential extension risk human activity index is obtained according to the abnormal degree. According to the causal correlation method, a city disaster chain semantic correlation network is constructed, multi-source data is fused to update the disaster evolution situation in real time, and disaster information and possible secondary disaster prediction are output. 2.The urban disaster risk assessment method according to claim 1, characterized in that: The hydrological information includes rainfall data obtained from the meteorological department, and river upstream and downstream water level, lake and reservoir water level and their warning information obtained from the water conservancy department.

3. The method of claim 1, wherein: The information about rainstorm in social media is obtained and disaster risk information is extracted therefrom, including: A set of risk keywords R = {w1, w2,..., w n} is extracted from the information about heavy rain in social media, different risk keywords correspond to different risk levels, and the potential expansion risk index PRL is defined as: The risk level of the keyword with the highest risk level is taken as the potential extension risk index of the whole set, and the maximum value of the potential extension risk index in the grid is taken as the potential extension risk level of the grid.

4. The method of claim 1, wherein the method further comprises: Abnormal information in the traffic speed data of the city mobile signaling and the floating car GPS before and after the rainstorm is obtained and quantified, and a potential extension risk human activity index is obtained according to the abnormal degree. Mobile signaling data and traffic speed data at the same location are obtained, and joint confidence is calculated, P l = P x *(1 + a*P j ) where P l is the joint confidence of the grid or road, indicating the likelihood of potential expansion risk at that location, P x is the abnormal probability of mobile phone signaling data, P j is the abnormal probability of vehicle speed data, and a is the population abnormality weight factor.

5. The method according to claim 4, wherein: The information reflected by the mobile signaling anomaly and the traffic anomaly is converted into a text description, and an association relationship is established with the information about rainstorm in social media according to the location and time, forming a merged text; A pre-trained large language model is used to extract causal correlation from the merged text; A triple city disaster risk propagation knowledge framework containing risk type-conduction relationship-propagation probability is constructed, and each node propagation probability includes disaster-causing factor X 致灾 and vulnerability parameter X 易损 wherein the risk type is the potential extension risk level, the propagation probability is based on the propagation probability in the historical cases of the grid in the region mined by a logistic regression model, and is obtained by the following formula: wherein β0is a constant term, β i , ε k is a regression coefficient.

6. A city disaster risk research and judgment device, comprising: An early warning analysis unit obtains a meteorological warning area according to meteorological warning information, obtains a flood warning area according to water regime warning, and spatially superimposes the meteorological warning area and the flood warning area to determine the urban area required for research and judgment; A multi-modal semantic information processing unit constructs a city disaster risk keyword library, distinguishes rainstorm-induced flood, geological disaster and secondary disaster risk, constructs a large language model risk research and judgment prompt word, obtains information about rainstorm in social media and extracts disaster risk information therefrom, and obtains a potential extension risk semantic index according to the disaster risk information; A city dynamic anomaly detection unit obtains abnormal information in the traffic speed data of the city mobile signaling and the floating car GPS before and after the rainstorm and quantifies it, and obtains a potential extension risk human activity index according to the abnormal degree. A disaster chain deduction unit constructs a city disaster chain semantic correlation network according to the causal correlation method, fuses multi-source data to update the disaster evolution situation in real time, and outputs disaster risk information and possible secondary disaster prediction.

Images