Urban disaster chain reasoning prediction method based on knowledge graph

Abstract

The invention discloses an urban disaster chain reasoning prediction method based on a knowledge graph, and relates to the technical field of intelligent disaster prevention and reduction. The method comprises the following steps: taking an initial disaster event node as a starting point, executing a weighted random walk algorithm on a dynamic probability causal graph, counting all accessed paths and path occurrence frequencies, and generating a high-probability disaster chain path set through normalization processing; performing intervention analysis on each high-probability disaster link in the high-probability disaster link path set through anti-factual reasoning logic to generate a key node analysis result; on the basis of the high-probability disaster chain path set and the key node analysis result, priority protection disaster-bearing bodies are screened out according to vulnerability sorting, risk reduction expected values are attached, and a disaster chain deduction report and a decision suggestion list are generated. According to the invention, the intelligence and decision reliability of urban disaster response are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent disaster prevention and mitigation technology, and in particular to a knowledge graph-based method for predicting urban disaster chains. Background Technology

[0002] With the continuous advancement of urbanization, the complexity and vulnerability of urban systems have significantly increased, and various disasters such as earthquakes, rainstorms, floods, and fires exhibit obvious spatiotemporal correlations and chain-like triggering effects. In recent years, research on urban disaster chains has gradually become an important direction in the field of disaster prevention and mitigation. Existing research mainly relies on multi-source heterogeneous data fusion, event logic reasoning, and statistical modeling to analyze the mechanisms of disaster occurrence and their propagation paths. For example, some methods use spatiotemporal data to construct disaster event networks and reveal the correlation between disasters through interaction models between nodes; other studies employ machine learning and big data analysis techniques to extract features and identify patterns from historical disaster data, thereby enabling the prediction and risk assessment of disaster chains. These technologies have played an important role in the identification and simulation of urban disaster chains, providing data support and theoretical basis for emergency decision-making.

[0003] Existing methods for disaster chain reasoning and prediction generally suffer from insufficient dynamic causal relationship updates and limited reasoning accuracy. Because disaster processes are influenced by multiple dynamic factors, traditional models based on static causal networks struggle to reflect the real-time evolution of urban disaster chains. Furthermore, disaster chain reasoning often remains at the level of event correlation analysis, lacking interpretability and intervention mechanisms based on causal reasoning, resulting in insufficient targeting of disaster prevention and control decisions. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a knowledge graph-based urban disaster chain reasoning and prediction method to solve the issues of difficulty in dynamically updating disaster causal relationships and insufficient interpretability of disaster chain reasoning results.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A knowledge graph-based urban disaster chain reasoning and prediction method, comprising,

[0007] Collect static and dynamic data, and based on the spatiotemporal causal ontology framework, convert the static and dynamic data into knowledge graph entities, attributes and relationships to generate an initial knowledge graph;

[0008] A conditional probability function is attached to each relation edge in the initial knowledge graph, and the conditional probability value is dynamically calibrated using a Bayesian update mechanism to generate a dynamic probabilistic causal graph.

[0009] Starting from the initial disaster event node, a weighted random walk algorithm is executed on the dynamic probabilistic causal graph to count all visited paths and their frequency of occurrence. After normalization, a set of high-probability disaster chain paths is generated.

[0010] By using counterfactual reasoning logic, intervention analysis is performed on each high-probability disaster link in the set of high-probability disaster chain paths to generate key node analysis results;

[0011] Based on the analysis results of high-probability disaster chain path sets and key nodes, priority disaster-bearing bodies are selected according to vulnerability ranking, and expected risk reduction values ​​are attached to generate a disaster chain simulation report and a list of decision-making recommendations.

[0012] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the static data includes urban basic geographic information, geological and geomorphological features, building structural parameters, infrastructure distribution, historical disaster records, and population socio-economic attributes.

[0013] The dynamic data includes meteorological monitoring data, hydrological data, traffic operation status, emergency response information, real-time sensor observation data, and social media disaster reports.

[0014] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the step of converting static and dynamic data into knowledge graph entities, attributes, and relationships based on a spatiotemporal causal ontology framework to generate an initial knowledge graph involves the following specific steps:

[0015] Based on a predefined spatiotemporal causal ontology framework, static and dynamic data are converted into initial knowledge graph entities, attributes, and relationships through named entity recognition and entity alignment algorithms.

[0016] Organize the entities, attributes, and relationships of the knowledge graph, and construct the initial knowledge graph using graph database methods.

[0017] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the following steps are taken: A conditional probability function is attached to each relation edge in the initial knowledge graph, and the conditional probability value is dynamically calibrated using a Bayesian update mechanism.

[0018] A dynamically variable conditional probability function is attached to each relation edge in the initial knowledge graph, and the historical observation count and success count of the relation edge are subjected to time decay processing to obtain new observation samples;

[0019] The new observation samples are weighted according to their confidence level and fused with the historical disaster event observation data in the initial knowledge graph after attenuation processing. The posterior parameters are recalculated according to the Bayesian update mechanism, and the conditional probability values ​​of the relation edges are dynamically calibrated.

[0020] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the dynamic probabilistic causal graph is obtained by completing the dynamic calibration of the conditional probabilities of each relation edge, writing the updated conditional probability values ​​as edge weights into the initial knowledge graph structure, and uniformly normalizing and verifying the connectivity of all nodes and relation edges, and storing and indexing the edge weights hierarchically according to time tags and spatial associations.

[0021] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the step of executing a weighted random walk algorithm on a dynamic probabilistic causal graph, starting from an initial disaster event node, and statistically analyzing all visited paths and their frequency of occurrence, is as follows:

[0022] Anomaly identification and event trigger determination are performed on the real-time data stream of the urban monitoring network to detect the initial disaster event node;

[0023] Starting from the initial disaster event node, based on the causal direction and weight distribution among the initial disaster event nodes, all initial disaster event nodes and their weight information in the dynamic probabilistic causal graph are loaded into memory. The initial disaster event node is selected as the starting point of the random walk. After completing all walks, the visited path is obtained.

[0024] Normalize each visited path, traverse all visited paths, calculate the average raw score and experience frequency of each visited path, and synthesize the path occurrence frequency by weighting.

[0025] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the high-probability disaster chain path set is generated by combining the path occurrence frequency of the initial disaster event node with the node conditional probability and edge weight, calculating the path score and normalizing it to a standard interval by linear normalization, and setting a dynamic threshold for filtering.

[0026] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the step of intervening and analyzing each high-probability disaster chain path in the set of high-probability disaster chain paths through counterfactual reasoning logic includes the following specific steps.

[0027] Based on the set of high-probability disaster chain paths, intervention hypotheses for target nodes are set, and the protection and response measures information of the corresponding nodes in the dynamic probability causal graph are called as intervention measures. The actual state of the target nodes in the original high-probability disaster chain path is replaced with the hypothetical state to generate an intervention hypothesis path set.

[0028] Based on the structural relationships and conditional probability distribution of the dynamic probabilistic causal graph, the causal propagation impact of the intervention state on the set of intervention hypothesis paths is calculated, and a post-propagation path probability set is generated.

[0029] Based on the post-propagation path probability set, compare the changes in path probability and risk differences before and after intervention, calculate the risk reduction effect and benefit indicators of each intervention measure, and output the intervention priority ranking results.

[0030] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of the present invention, the key node analysis results are based on the set of high-probability disaster chain paths and the intervention priority ranking results. High-priority intervention target nodes are selected and relevant parameters are collected. The outbound edge conditional probability of the intervened nodes is updated according to the closed rule. The updated results are mapped to the affected subgraph. Weighted random walk reasoning is performed on the affected subgraph and generated by statistical path probability distribution.

[0031] As a preferred embodiment of the knowledge graph-based urban disaster chain reasoning and prediction method of this invention, the steps include: based on the high-probability disaster chain path set and key node analysis results, selecting priority disaster-bearing bodies according to vulnerability ranking, attaching expected risk reduction values, and generating a disaster chain projection report and decision recommendation list. The specific steps are as follows.

[0032] Extract the disaster-bearing body attribute set of each disaster-bearing body from the high-probability disaster chain path set, and spatially match the disaster-bearing body attribute set with the influence range of key nodes recorded in the key node analysis results to obtain the corresponding relationship of disaster-bearing body influence.

[0033] Based on the correspondence between disaster-bearing bodies and their impacts, the multi-index entropy weight-TOPSIS method is used to standardize the attribute set of disaster-bearing bodies and determine their weights. The relative proximity of each disaster-bearing body is calculated to generate a comprehensive risk potential value.

[0034] Based on the comprehensive risk potential value and the corresponding relationship between the impact of disaster-bearing bodies, the expected risk reduction value of each disaster-bearing body is calculated.

[0035] Based on the expected risk reduction value, the vulnerability of disaster-bearing bodies is ranked, and the key disaster-bearing body set is selected through priority protection threshold. The corresponding expected risk reduction value and protection recommendations are output, and a disaster chain simulation report and decision list are generated.

[0036] The beneficial effects of this invention are as follows: By embedding conditional probability functions into the knowledge graph and employing Bayesian mechanisms for dynamic calibration, real-time updates of disaster causal relationships are achieved, enabling the graph to adaptively evolve with environmental changes. Simultaneously, by combining counterfactual reasoning logic, virtual intervention analysis is conducted on different protective measures, quantifying the risk mitigation effect of each node. This enhances the accuracy and timeliness of disaster chain reasoning and prediction, provides a scientific basis for optimizing disaster prevention and control plans, and improves the intelligence and reliability of urban disaster response decisions. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0038] Figure 1 This is a flowchart of a knowledge graph-based urban disaster chain reasoning and prediction method.

[0039] Figure 2 This is a flowchart of the initial knowledge graph.

[0040] Figure 3 This is a flowchart for dynamic calibration of conditional probability.

[0041] Figure 4 A flowchart for prioritizing the protection and reporting of disaster-bearing entities. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0045] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a knowledge graph-based urban disaster chain reasoning and prediction method, including the following steps:

[0046] S1. Collect static and dynamic data, and based on the spatiotemporal causal ontology framework, convert the static and dynamic data into knowledge graph entities, attributes and relationships to generate an initial knowledge graph.

[0047] It should be noted that static data includes basic urban geographic information, geological and geomorphological features, building structural parameters, infrastructure distribution, historical disaster records, and population socioeconomic attributes.

[0048] Dynamic data includes meteorological monitoring data, hydrological data, traffic operation status, emergency response information, real-time sensor observation data, and social media disaster reports.

[0049] S1.1 Based on the predefined spatiotemporal causal ontology framework, static and dynamic data are converted into initial knowledge graph entities, attributes, and relationships through named entity recognition and entity alignment algorithms.

[0050] Furthermore, the input data is formatted and preprocessed, unified into a structured format, and the timestamps are corrected to ISO8601. Time and location are labeled using a time index table and a spatial reference grid, resulting in standardized data packets and preprocessed logs. Subsequently, a named entity recognition algorithm is applied to the standardized data to extract candidate entities and generate a candidate entity list. Entities are normalized and disambiguated to obtain unique entity IDs and form an entity identifier table. At the same time, static and dynamic attributes are extracted according to a predefined attribute vocabulary and an attribute dictionary is constructed, generating an entity identifier table and an attribute dictionary. Based on the entity identifier table and attribute dictionary, entities and attributes are mapped to spatiotemporal causal ontology concepts. Spatiotemporal delimitation is performed using time indexes and spatial grids to resolve conflicts and assign confidence scores, ultimately generating an initial knowledge graph.

[0051] It should be noted that the spatiotemporal causal ontology framework is built on the basis of integrating the semantic relationships of multi-source disaster data. It establishes a conceptual hierarchy and attribute system by aggregating concepts such as disaster events, disaster-bearing bodies, environmental elements and response behaviors; it clarifies the spatiotemporal constraints and dependencies of each concept based on the time index table and spatial reference grid, and forms the causal chain between concepts by combining historical causal samples.

[0052] S1.2 Organize the entities, attributes, and relationships of the knowledge graph, and construct the initial knowledge graph using graph database methods.

[0053] Furthermore, consistency checks and deduplication are performed on the entity identifier table, attribute dictionary, and relation list. Attribute names and units are standardized based on the attribute dictionary, and time index tables and spatial reference grid information are completed for each relation. Subsequently, the verified entity-attribute-relationship is converted into a standard triple format, generating a graph database import package containing node type, edge type, attribute fields, entity ID index, and time-space index. Then, the triple dataset is batch-processed, entity ID index and time index are created, and connectivity and confidence integrity checks are performed. Versioned snapshots and import logs are recorded, ultimately producing a queryable, indexed, and versioned initial knowledge graph.

[0054] S2. Append a conditional probability function to each relation edge in the initial knowledge graph, and use the Bayesian update mechanism to dynamically calibrate the conditional probability values ​​to generate a dynamic probabilistic causal graph.

[0055] S2.1. Append a conditional probability function to each relation edge in the initial knowledge graph, and dynamically calibrate the conditional probability value using a Bayesian update mechanism.

[0056] Furthermore, based on the relationship type and historical statistics, a prior probability distribution is set for each relation edge. The real-time and historical observations are mapped to corresponding relations using a time index table and a spatial reference grid to calculate the observation likelihood. The prior is gradually corrected according to the Bayesian update formula and the observation likelihood. The fusion of multi-source observations and spatiotemporal windowing updates are processed. The results of each update are normalized and versioned to obtain the conditional probability value.

[0057] S2.2 The dynamic probabilistic causal graph is obtained by dynamically calibrating the conditional probabilities of each relation edge, writing the updated conditional probability values ​​as edge weights into the initial knowledge graph structure, uniformly normalizing and verifying the connectivity of all nodes and relation edges, and storing and indexing the edge weights hierarchically according to time tags and spatial associations.

[0058] Furthermore, the updated conditional probability values ​​are written into the initial knowledge graph structure as edge weights. The overall structure is uniformly normalized and its connectivity is verified based on the node relationships to ensure that the probability distribution is consistent and the relationships are complete. Combined with time tags and spatial relationships, the edge weights are stored and indexed in a hierarchical manner, ultimately forming a dynamic probabilistic causal graph that can be dynamically queried and reasoned about in terms of spatiotemporal dimensions.

[0059] It should be noted that the time tag represents the specific time information corresponding to each relationship or edge, which usually comes from the time index table during data preprocessing and is used to distinguish the state changes at different points in time or time periods.

[0060] Spatial associations represent the spatial location or extent information involved in a relationship or edge. They can be labeled using a spatial reference grid to mark the geographical or spatial location of entities or relationships, supporting spatiotemporal analysis and index management.

[0061] Time tags identify the time dimension of a relationship, while spatial associations identify the spatial location or range of a relationship. The combination of the two is used for spatiotemporal dynamic querying and reasoning.

[0062] S3. Starting from the initial disaster event node, execute a weighted random walk algorithm on the dynamic probabilistic causal graph, count all visited paths and their frequency of occurrence, and generate a set of high-probability disaster chain paths through normalization.

[0063] S3.1. Perform anomaly identification and event trigger determination on the real-time data stream of the urban monitoring network to detect the initial disaster event node.

[0064] Furthermore, based on a dynamic probabilistic causal graph, real-time data streams from the urban monitoring network are accessed, and after unifying the format and timestamps, spatiotemporal registration is performed according to a time index table and spatial reference grid. Missing value imputation and noise filtering are performed on the observation data, key statistical features within the sliding time window are extracted, and spatiotemporal feature vectors are generated. Then, the observation consistency and anomaly score are calculated by combining the dynamic probabilistic causal graph, and anomalies are identified and multi-source results are fused based on Bayesian factors and anomaly score judgment criteria. Next, anomaly alarms are clustered in the time and space dimensions to form candidate events, and the type and severity are evaluated using historical event templates and causal relationships. Finally, the initial disaster event nodes are determined, and the initial disaster event node set is output.

[0065] S3.2 Starting from the initial disaster event node, based on the causal direction and weight distribution among the initial disaster event nodes, load all the initial disaster event nodes and weight information in the dynamic probabilistic causal graph into memory, select the initial disaster event node as the starting point of the random walk, and obtain the visited path after completing all walks.

[0066] Furthermore, the dynamic probabilistic causal graph, including causal direction and edge weight distribution, is loaded into memory as a whole, and all initial disaster event nodes and their corresponding edge weights are extracted. The edge weights are normalized according to the conditional probability values ​​to construct the transition probability distribution starting from each node. Then, the walk step size, number of times, and backtrack probability are set with the initial disaster event node as the starting point, and random walks are performed according to the transition probability and causal direction. During the walk, the spatiotemporal rationality of the path is constrained by the time index table and spatial reference grid. After all walks are completed, the number of visits and path weights are counted, duplicate paths are merged and filtered, and finally, the set of visited paths verified by spatiotemporal constraints is output.

[0067] S3.3. Normalize each visited path, traverse all visited paths, calculate the average original score and experience frequency of each visited path, and synthesize the path occurrence frequency by weighting.

[0068] Furthermore, normalization is performed on each visited path. The original edge weights of each edge in the path are normalized according to the path length and the sum of edge weights, while retaining the time label and spatial range, to generate a normalized path table. The normalized path table and the path visit count table are traversed, and the average original score (weighted average of edge weights or path scores before normalization) and empirical frequency (path visit count divided by the total number of all random walks) are calculated for each path. Then, the average original score and empirical frequency are weighted and synthesized according to a preset composite weight coefficient determined based on historical data statistics or experience, to obtain the weighted occurrence frequency of each path.

[0069] S4. Through counterfactual reasoning logic, intervene and analyze each high-probability disaster link in the set of high-probability disaster chain paths to generate key node analysis results.

[0070] S4.1 Based on the set of high-probability disaster chain paths, set the intervention hypothesis for the target node, and call the protection and response measures information of the corresponding node in the dynamic probability causal graph as the intervention measures. Replace the actual state of the target node in the original high-probability disaster chain path with the hypothetical state to generate the intervention hypothesis path set.

[0071] Furthermore, based on the set of high-probability disaster chain paths, one or more target nodes are selected from each path, and intervention hypotheses are defined (e.g., the state is changed to "controlled" or "isolated"). The protection and response measures information of the corresponding target nodes are read from the dynamic probabilistic causal graph, and these measures are mapped to the hypothetical state change rules formed by the change of node state. Then, according to the hypothetical state change rules, the actual state of each target node in the original high-probability disaster chain path is replaced with the hypothetical state, and the time label and spatial association of the node are updated to maintain spatiotemporal consistency. For each replaced path, the path probability is recalculated based on the conditional probability function of the edge, and the probability difference and confidence level before and after the change are recorded. Feasibility verification is performed, and hypothetical paths that do not meet the time or resource constraints are eliminated. Finally, the paths that pass the verification are summarized, deduplicated, and versioned for storage, generating an intervention hypothetical path set with hypothetical state and updated probability information.

[0072] S4.2 Based on the structural relationship and conditional probability distribution of the dynamic probability causal graph, calculate the causal propagation impact of the intervention state on the intervention hypothesis path set, and generate the post-propagation path probability set.

[0073] Furthermore, by using an intervention hypothesis path set with assumed states and updated probability information, the nodes and edges in the relevant paths are aligned in memory with the topology and conditional probability distribution of the dynamic probabilistic causal graph, clarifying the time labels and spatial associations involved in each path. Probability propagation is performed sequentially for each hypothesis path in chronological order: using the assumed states of nodes as observation conditions, Bayesian probability transfer and marginalization operations are performed along the causal direction using the conditional probability functions of edges. When encountering loops, iterative belief propagation or variational inference is used to eliminate dependencies and maintain convergence. During the propagation process, the temporal propagation delay and spatial influence range are corrected by combining a time index table and a spatial reference grid, and the influence of shared nodes among multiple paths is coupled and calculated using joint marginalization or conditional decomposition. After the propagation is completed, the overall probability of each path is recalculated based on the posterior probability of nodes and the conditional probability of edges, and normalized and labeled with confidence. The probability differences before and after propagation and versioned snapshots are recorded, and finally, the path probability set after propagation is output.

[0074] S4.3 Based on the post-propagation path probability set, compare the changes in path probability and risk differences before and after intervention, calculate the risk reduction effect and benefit indicators of each intervention measure, and output the intervention priority ranking results.

[0075] Furthermore, based on the post-propagation path probability set, the system calls upon the probability snapshots and path metadata recorded in the intervention hypothesis path set before and after the change, calculates the probability change value for each path (the difference between the probability before the change and the probability after propagation is used as the value), and reads the impact metric (e.g., the product of exposure, vulnerability, and asset value) corresponding to each path from the knowledge graph or external loss estimation table to convert the probability change into an absolute risk reduction value. The risk reduction value of each path is back-summarized to each intervention measure according to its corresponding target node (the intervention cost is obtained by using the protection and response information of the nodes in the dynamic probability causal graph). The system calculates the total risk reduction effect, unit cost-benefit ratio, and confidence-weighted benefit based on path confidence for each intervention. Then, all intervention measures are ranked according to multiple indicators based on the total risk reduction effect, unit cost-benefit ratio, and confidence-weighted score, and constraints are selected by combining available resources, time window, and implementation feasibility. Finally, a list of intervention measures ranked by priority is output.

[0076] S4.4 Based on the set of high-probability disaster chain paths and the results of intervention priority ranking, select high-priority intervention target nodes and collect relevant parameters. Update the outbound edge conditional probability of the intervened nodes according to the closed rule, map the update results to the affected subgraph, perform weighted random walk inference on the affected subgraph, and generate key node analysis results by statistical path probability distribution.

[0077] Furthermore, following the output of the high-probability disaster chain path set and intervention priority ranking results, key nodes in the paths are selected from high to low intervention priority. High-priority intervention target nodes are selected, and their corresponding operational parameters, environmental data, and status indicators are collected for subsequent intervention analysis and probability updates (such as node current status, time label, spatial reference grid identifier, and protection and response measures attributes). Then, based on closed-form rules (obtained by analyzing the causal relationships and event trigger probabilities between nodes in historical disaster data and establishing mathematical expressions using Bayes' theorem), the outbound edge conditional probability of each intervened node is updated using the collected parameters, and the updated edge weights are mapped to the corresponding affected subgraph to construct the latest probability structure of the affected subgraph. Finally, weighted random walk inference is performed on the affected subgraph to statistically analyze the path probability distribution and calculate the frequency of node occurrence, path accessibility, and probability changes under intervention conditions, outputting key node analysis results to guide decision-making.

[0078] S5. Based on the analysis results of the set of high-probability disaster chain paths and key nodes, select priority disaster-bearing bodies according to vulnerability ranking, attach the expected risk reduction value, and generate a disaster chain simulation report and a list of decision recommendations.

[0079] S5.1 Extract the disaster-bearing body attribute set of each disaster-bearing body from the high-probability disaster chain path set, and perform spatial correlation matching between the disaster-bearing body attribute set and the influence range of key nodes recorded in the key node analysis results to obtain the corresponding relationship of disaster-bearing body influence.

[0080] Furthermore, a list of affected entities is extracted from each path, and an attribute set (including location coordinates, spatial boundaries, exposure, vulnerability, protection level, importance / asset value, and time-related status) is constructed for each affected entity. The attribute set is then output as a unique identifier table for the affected entity. The influence range of key nodes recorded in the key node analysis results (determined based on the outbound conditional probability and path probability distribution of key nodes in the affected subgraph, reflecting the spatiotemporal effect area of ​​key nodes in the dynamic probabilistic causal graph) and spatial reference grid are loaded. Spatiotemporal registration is performed between the affected entity location and the influence range of key nodes based on the time index table, and spatial intersection and coverage are calculated. Then, based on attributes such as coverage, exposure, and vulnerability, the influence score and confidence level are calculated for each pair of affected entity-key node influence ranges, and weighted allocation is performed according to the overlapping area or exposure weight when multiple coverages exist. Finally, all matching pairs are summarized into a table of affected entity influence correspondence, recording the affected entity ID, key node ID, spatial overlap index, influence score, confidence level, and corresponding time window, and the affected entity influence correspondence is output.

[0081] S5.2 Based on the correspondence of the impact of disaster-bearing bodies, the multi-index entropy weight-TOPSIS method is used to standardize the attribute set of disaster-bearing bodies and determine the weights, calculate the relative proximity of each disaster-bearing body, and generate a comprehensive risk potential value.

[0082] Furthermore, based on the disaster-bearing body IDs recorded in the disaster-bearing body impact correspondence table, a disaster-bearing body attribute set is summarized, and a multi-indicator matrix including indicators such as exposure, vulnerability, protection cost, and importance is constructed. The range standardization method is used to unify the values ​​of each indicator to the [0,1] interval to form a standardized matrix. Based on the information entropy of each indicator in the disaster-bearing body sample, the information entropy value and redundancy are calculated, and then the entropy weight of each indicator is obtained and an entropy weight vector is established. Then, based on the standardized matrix, the Euclidean distance from each disaster-bearing body to the ideal optimal solution and the ideal worst solution is calculated, and the relative proximity is used as a comprehensive risk evaluation indicator. Finally, the relative proximity value of each disaster-bearing body is normalized and used as the comprehensive risk potential value, and the comprehensive risk potential result table of disaster-bearing bodies is output.

[0083] S5.3. Based on the comprehensive risk potential value and the corresponding relationship between the impact of disaster-bearing bodies, calculate the expected risk reduction value of each disaster-bearing body.

[0084] Furthermore, the analysis of the comprehensive risk potential energy result table of disaster-bearing bodies and the corresponding relationship table of disaster-bearing body impacts is performed. For each available intervention measure, the protection and response measures information of the corresponding node in the dynamic probability causal graph are matched, and their effect coefficients and implementation coverage rates are read. Based on the corresponding relationship table of disaster-bearing body impacts, the spatial coverage and temporal consistency of each intervention for each disaster-bearing body are determined. For each disaster-bearing body, the risk reduction contribution value is calculated item by item according to the formula, and the contributions from multiple interventions to the same disaster-bearing body are weighted and summed to obtain the total expected risk reduction value for that disaster-bearing body. Then, the expected risk reduction value is normalized and subjected to uncertainty sensitivity analysis (recalculated through different confidence level scenarios). The results are sorted by the absolute value of risk reduction and unit cost-benefit, and versioned snapshots and source proofs are recorded to generate a table of expected risk reduction values ​​for disaster-bearing bodies.

[0085] S5.4. Based on the expected risk reduction value, the vulnerability of disaster-bearing bodies is ranked, and the key disaster-bearing body set is selected through the priority protection threshold. The corresponding expected risk reduction value and protection suggestions are output, and a disaster chain simulation report and decision list are generated.

[0086] Furthermore, a vulnerability score is calculated for each disaster-bearing entity in the table based on its expected risk reduction value and its uncertainty range (the comprehensive risk potential value, the reciprocal of the expected risk reduction value, and the exposure are incorporated into the weighted scoring function), and a vulnerability ranking list is generated accordingly. Then, the vulnerability ranking list is filtered based on pre-set priority protection thresholds and available resource constraints to select a set of key disaster-bearing entities, and the selection criteria and threshold parameters are recorded. Next, for each key disaster-bearing entity, corresponding protection and response measures information is retrieved from the knowledge graph, and targeted protection recommendations and implementation plan drafts are generated by combining implementation costs, coverage, and time windows. Subsequently, cost-benefit sensitivity analysis is performed on the recommendations, and priority robustness is verified under different confidence level scenarios, revising the ranking and recommendations accordingly. Finally, the vulnerability ranking, key disaster-bearing entity set, expected risk reduction value, protection recommendations, cost estimation, and verification results are compiled into a structured disaster chain simulation report, and a decision list that can be directly used for scheduling is exported, resulting in the disaster chain simulation report and decision list.

[0087] It should be noted that the priority protection threshold is set based on the statistical distribution of vulnerability scores of disaster-bearing bodies and the historical disaster loss levels, and is used to classify high, medium, and low priority protection targets. First, the vulnerability scores of all disaster-bearing bodies are standardized, and their mean and standard deviation are calculated. Then, segmented thresholds are set based on the distribution locations of high-risk disaster-bearing bodies in historical data.

[0088] The high priority protection threshold is set based on the vulnerability distribution characteristics of high-loss disaster-bearing bodies in historical disaster data, and is usually a scoring range of mean + 0.5 × standard deviation or above;

[0089] The priority protection threshold is set based on the statistical distribution of the vulnerability scores of all disaster-bearing entities, and is usually between the mean and the mean + 0.5 × standard deviation.

[0090] The low priority protection threshold is set based on the overall statistical distribution of the vulnerability score of the disaster-bearing body, which is usually the range of scores of all disaster-bearing bodies that are lower than the mean.

[0091] In summary, this invention achieves real-time updates of disaster causal relationships by embedding conditional probability functions into the knowledge graph and employing Bayesian mechanisms for dynamic calibration, enabling the graph to adaptively evolve with environmental changes. Simultaneously, by combining counterfactual reasoning logic, it conducts virtual intervention analysis on different protective measures, quantifying the risk mitigation effect of each node. This enhances the accuracy and timeliness of disaster chain reasoning and prediction, provides a scientific basis for optimizing disaster prevention and control plans, and improves the intelligence and reliability of urban disaster response decisions.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A knowledge graph-based method for predicting urban disaster chains, characterized in that: include, Collect static and dynamic data, and based on the spatiotemporal causal ontology framework, convert the static and dynamic data into knowledge graph entities, attributes and relationships to generate an initial knowledge graph; A conditional probability function is attached to each relation edge in the initial knowledge graph, and the conditional probability value is dynamically calibrated using a Bayesian update mechanism to generate a dynamic probabilistic causal graph. Starting from the initial disaster event node, a weighted random walk algorithm is executed on the dynamic probabilistic causal graph to count all visited paths and their frequency of occurrence. After normalization, a set of high-probability disaster chain paths is generated. By using counterfactual reasoning logic, intervention analysis is performed on each high-probability disaster link in the set of high-probability disaster chain paths to generate key node analysis results; Based on the analysis results of high-probability disaster chain path sets and key nodes, priority disaster-bearing bodies are selected according to vulnerability ranking, and expected risk reduction values ​​are attached to generate a disaster chain simulation report and a list of decision-making recommendations.

2. The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 1, characterized in that: The static data includes urban basic geographic information, geological and geomorphological features, building structural parameters, infrastructure distribution, historical disaster records, and population socioeconomic attributes; The dynamic data includes meteorological monitoring data, hydrological data, traffic operation status, emergency response information, real-time sensor observation data, and social media disaster reports. 3.The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 2, characterized in that: Based on the spatiotemporal causal ontology framework, static and dynamic data are transformed into knowledge graph entities, attributes, and relationships to generate an initial knowledge graph. The specific steps are as follows. Based on a predefined spatiotemporal causal ontology framework, static and dynamic data are transformed into initial knowledge graph entities, attributes, and relationships through named entity recognition and entity alignment algorithms. Organize the entities, attributes, and relationships of the knowledge graph, and construct the initial knowledge graph using graph database methods. 4.The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 1, characterized in that: The specific steps for attaching a conditional probability function to each relation edge in the initial knowledge graph and dynamically calibrating the conditional probability value using a Bayesian update mechanism are as follows. A dynamically variable conditional probability function is attached to each relation edge in the initial knowledge graph, and the historical observation count and success count of the relation edge are subjected to time decay processing to obtain new observation samples; The new observation samples are weighted according to their confidence level and fused with the historical disaster event observation data in the initial knowledge graph after attenuation processing. The posterior parameters are recalculated according to the Bayesian update mechanism, and the conditional probability values ​​of the relation edges are dynamically calibrated. 5.The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 4, characterized in that: The dynamic probabilistic causal graph is obtained by dynamically calibrating the conditional probabilities of each relation edge, writing the updated conditional probability values ​​as edge weights into the initial knowledge graph structure, uniformly normalizing and verifying the connectivity of all nodes and relation edges, and storing and indexing the edge weights hierarchically according to time tags and spatial associations. 6.The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 1, wherein: Starting from the initial disaster event node, a weighted random walk algorithm is executed on the dynamic probabilistic causal graph to count all visited paths and their frequency of occurrence. The specific steps are as follows. Anomaly identification and event trigger determination are performed on the real-time data stream of the urban monitoring network to detect the initial disaster event node; Starting from the initial disaster event node, the causal direction and weight distribution among the initial disaster event nodes are considered; Load all initial disaster event nodes and their weight distributions from the dynamic probabilistic causal graph into memory, select the initial disaster event nodes as the starting points for random walks, and obtain the visited paths after completing all walks. Normalize each visited path, traverse all visited paths, calculate the average raw score and experience frequency of each visited path, and synthesize the path occurrence frequency by weighting.

7. The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 6, characterized in that: The high-probability disaster chain path set is generated by combining the path occurrence frequency of the initial disaster event node with the node conditional probability and edge weight, calculating the path score, and then linearly normalizing it to a standard interval, and setting a dynamic threshold for filtering. 8.The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 1, characterized in that: The method involves using counterfactual reasoning logic to analyze and intervene in each high-probability disaster link in the set of high-probability disaster chain paths. The specific steps are as follows. Based on the set of high-probability disaster chain paths, intervention hypotheses for target nodes are set, and the protection and response measures information of the corresponding nodes in the dynamic probability causal graph are called as intervention measures. The actual state of the target nodes in the original high-probability disaster chain path is replaced with the hypothetical state to generate an intervention hypothesis path set. Based on the structural relationships and conditional probability distribution of the dynamic probabilistic causal graph, the causal propagation impact of the intervention state on the set of intervention hypothesis paths is calculated, and a post-propagation path probability set is generated. Based on the post-propagation path probability set, compare the changes in path probability and risk differences before and after intervention, calculate the risk reduction effect and benefit indicators of each intervention measure, and output the intervention priority ranking results.

9. The knowledge graph-based urban disaster chain reasoning and prediction method as described in claim 8, characterized in that: The key node analysis results are based on the set of high-probability disaster chain paths and the intervention priority ranking results. High-priority intervention target nodes are selected and relevant parameters are collected. The outbound edge conditional probability of the intervened nodes is updated according to the closed rule. The updated results are mapped to the affected subgraph. Weighted random walk inference is performed on the affected subgraph and generated through statistical path probability distribution. 10.The knowledge graph-based urban disaster chain reasoning and prediction method according to claim 1, characterized in that: Based on the analysis results of high-probability disaster chain paths and key nodes, priority disaster-bearing bodies are selected according to vulnerability ranking, and expected risk reduction values ​​are attached. A disaster chain projection report and a list of decision-making recommendations are then generated. The specific steps are as follows: Extract the disaster-bearing body attribute set of each disaster-bearing body from the high-probability disaster chain path set, and spatially match the disaster-bearing body attribute set with the influence range of key nodes recorded in the key node analysis results to obtain the corresponding relationship of disaster-bearing body influence. Based on the correspondence between disaster-bearing bodies and their impacts, the multi-index entropy weight-TOPSIS method is used to standardize the attribute set of disaster-bearing bodies and determine their weights. The relative proximity of each disaster-bearing body is calculated to generate a comprehensive risk potential value. Based on the comprehensive risk potential value and the corresponding relationship between the impact of disaster-bearing bodies, the expected risk reduction value of each disaster-bearing body is calculated. Based on the expected risk reduction value, the vulnerability of disaster-bearing bodies is ranked, and a set of key disaster-bearing bodies is selected through priority protection thresholds. The corresponding expected risk reduction value and protection recommendations are output, and a disaster chain simulation report and decision list are generated.

Images