Bayesian network-based urban disaster reasoning and risk assessment method and system

Abstract

This invention provides a method and system for urban disaster reasoning and risk assessment based on Bayesian networks, belonging to the field of emergency management. The method includes: acquiring multi-source real-time monitoring data, including urban infrastructure and meteorological data; acquiring historical event data and constructing a Bayesian network based on the historical event data, wherein the Bayesian network includes disaster-causing factor nodes, intermediate event nodes, and disaster outcome nodes; inputting real-time monitoring data as evidence into the Bayesian network model to update the state of relevant nodes in the Bayesian network; performing conditional probability reasoning based on the updated Bayesian network to calculate the probability of occurrence of power outage events and related impact events; and classifying and assessing the power outage risk based on the probability of occurrence to obtain risk level and corresponding disaster chain evolution path information, thereby achieving causal analysis, risk prediction, and impact assessment of power outage disasters.

Description

Technical Field

[0001] This invention relates to the field of urban safety and emergency management technology, and more specifically, to a method and system for urban disaster reasoning and risk assessment based on Bayesian networks. Background Technology

[0002] Under extreme weather conditions (such as heavy rainfall and flooding) or external disturbances, urban power systems are highly coupled with infrastructure such as transportation, communication, and drainage, which can easily lead to complex disaster events centered on "large-scale power outages".

[0003] Existing methods for assessing power outage risks mainly rely on threshold judgments, empirical rules, or single-model predictions, which have the following shortcomings: 1. It is difficult to depict the causal relationships between multiple disaster-causing factors; 2. It lacks the ability to systematically analyze the causal chain and evolution path of power outage events; 3. Risk assessment results are mostly static outputs and are difficult to update dynamically with real-time data; 4. The reasoning process is not transparent, which is not conducive to decision support in emergency management scenarios.

[0004] This invention aims to solve the above problems by proposing a technical solution for dynamic reasoning and risk assessment of urban large-scale power outage disaster chains based on Bayesian networks, which enables the analysis of the causes of power outage disasters, risk prediction, and impact assessment. Summary of the Invention

[0005] In view of this, the present invention proposes a method and system for urban disaster reasoning and risk assessment based on Bayesian networks to solve the problems existing in the prior art.

[0006] To achieve the above objectives, this invention proposes a method and system for urban disaster reasoning and risk assessment based on Bayesian networks, comprising: Acquire real-time monitoring data from multiple sources, including urban infrastructure and meteorological data; Acquire historical event data and construct a Bayesian network based on the historical event data, wherein the Bayesian network includes disaster-causing factor nodes, intermediate event nodes, and disaster result nodes; The real-time monitoring data is used as evidence to input into the Bayesian network model, and the state of the relevant nodes in the Bayesian network is updated. Conditional probability inference is performed based on the updated Bayesian network to calculate the probability of occurrence of power outage events and related impact events; The risk of power outages is classified and assessed based on the probability of occurrence, resulting in information on the risk level and the corresponding disaster chain evolution path.

[0007] Optionally, the real-time monitoring data includes power grid operation data, meteorological data, hydrological data, and urban infrastructure operation status data.

[0008] Optionally, the process of constructing a Bayesian network includes: Based on historical event data, corresponding disaster-causing factor nodes, intermediate event nodes, and disaster outcome nodes are assigned. Conditional independence is tested using conditional mutual information based on historical event data, and a network skeleton is formed for different nodes. The network skeleton is optimized based on temporal constraint mechanism and scoring function optimization, and prior probabilities are set for different nodes to generate a Bayesian network.

[0009] Optionally, the process of inputting the real-time monitoring data as evidence into the Bayesian network model includes: The real-time monitoring data is transformed to generate corresponding state probabilities. Hard evidence is injected into the direct observation results in the real-time monitoring data, and soft evidence is injected into the uncertain data in the real-time monitoring data to generate corresponding probability distribution evidence. After injection, the prior probabilities of the nodes of the corresponding Bayesian network are replaced with probability distribution evidence.

[0010] Optionally, the process of updating the state of the relevant nodes in the Bayesian network includes: Based on the input evidence, the belief distribution of all nodes in the Bayesian network is updated using a message passing algorithm. The message passing algorithm includes: passing causal information from parent nodes to child nodes along directed edges of the network via a first message, and passing diagnostic information from child nodes to parent nodes along directed edges of the network via a second message. Each node calculates its updated belief distribution as the posterior probability of its node state in the Bayesian network based on the received first and second messages and its own conditional probability table.

[0011] Optionally, the calculation process for the probability of power outage events and related impact events includes: Based on the posterior probability distribution of each node in the updated Bayesian network, the probability of a power outage event is calculated by probability summation and conditional probability propagation. Specifically, for a power outage event defined as a single node in a specific state, its probability of occurrence is directly read from the posterior probability of the corresponding state of that node; for a cascading power outage event defined as a logical combination of the states of several nodes, its probability of occurrence is obtained by summing the posterior joint probabilities of all related state combinations of the several nodes. For secondary impact events of a power outage, the probability of occurrence of the secondary impact event is calculated using the total probability formula, which combines the posterior probability of the power system state and the conditional probability of the secondary impact event under a given power system state.

[0012] Optionally, the process of classifying and assessing power outage risks based on their probability of occurrence includes: The comprehensive risk index is calculated based on the probability of occurrence and the corresponding depth of the disaster chain propagation. The comprehensive risk index is then classified into different thresholds to obtain the risk level.

[0013] Optionally, the process of obtaining the disaster chain evolution path information includes: The propagation path of the Bayesian network nodes corresponding to the power outage event is evaluated, and the corresponding path score is generated. Based on the path score, the corresponding disaster chain evolution path information is selected.

[0014] On the other hand, the present invention provides a Bayesian network-based urban disaster reasoning and risk assessment system for performing the above-described method.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. A disaster chain modeling approach for large-scale urban power outage scenarios is proposed; 2. Utilize Bayesian networks to achieve causal reasoning and dynamic probability updates for power outage disasters; 3. Enable zoned and tiered assessment and output of power outage risks to directly support emergency response decision-making; 4. The reasoning process has clear causal logic, and the results have good interpretability. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention. Detailed Implementation

[0017] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] This invention proposes a method and system for urban disaster reasoning and risk assessment based on Bayesian networks, belonging to the field of urban safety and emergency management technology. Specifically, it relates to a method and system implementation for dynamic reasoning and risk assessment of disaster chains in the context of large-scale power outages in cities, which is applicable to power emergency response, urban disaster prevention and mitigation, and emergency decision support.

[0019] The core idea of ​​this invention is to treat large-scale urban television outages as a "disaster chain event" formed by the coupling of multiple factors, and to use Bayesian networks to probabilistically model the disaster chain, combining real-time data to achieve dynamic reasoning and risk assessment. Specifically: 1. Abstract factors such as meteorology, hydrology, power grid operation, and infrastructure vulnerability into nodes in a Bayesian network; 2. Construct a power outage disaster chain model by describing the conditional dependencies between nodes using directed edges; 3. Update the node status under real-time or near real-time data input conditions; 4. Calculate the probability of occurrence of power outage events and related impact events using probabilistic reasoning algorithms; 5. Output power outage risk assessment results by region and level, as well as the evolution path of the disaster chain.

[0020] This method emphasizes causal interpretability, dynamic updating capability, and adaptability to emergency decision-making.

[0021] The method provided by this invention is as follows: like Figure 1 As shown, the method of the present invention includes the following steps: 1. Multi-source data acquisition Acquire multi-source data related to urban power outages, including but not limited to: power grid operation data (load, voltage, equipment status, etc.); meteorological data (rainfall, wind, etc.); hydrological data (flooding, water level, etc.); urban geographic information and critical infrastructure operation status data.

[0022] The above-mentioned multi-source data were acquired through data management systems of different data types. These included: meteorological data (hourly rainfall, instantaneous wind speed, temperature, relative humidity, etc.) sourced from meteorological monitoring stations and numerical weather prediction models; hydrological data (water depth, river level, drainage network pressure, etc.) sourced from hydrological monitoring stations and urban flooding monitoring points; power grid operation data (node ​​voltage, line current, transformer load rate, circuit breaker status) sourced from SCADA systems and PMU devices; and infrastructure data (traffic flow, communication signal strength, power supply status of important users, etc.) sourced from urban operation monitoring platforms. Preprocessing of the above multi-source data included spatiotemporal alignment, filling missing data at different locations using spatial interpolation, and removing outliers and filling in missing values.

[0023] 2. Disaster Chain Bayesian Network Modeling Based on historical event data, a Bayesian network structure for urban power outage disaster chains is constructed, with nodes including disaster-causing factor nodes, intermediate event nodes, and disaster outcome nodes.

[0024] Specifically, disaster-causing factor nodes include meteorological nodes such as rainfall intensity and wind speed level, hydrological nodes such as water depth and river water level, and equipment nodes such as transformer load rate and insulation status. Intermediate event nodes include nodes such as substation water ingress risk, line galloping probability, equipment insulation degradation, and protection malfunction risk. Disaster outcome nodes include nodes such as power outage range, power outage duration, power loss ratio of important users, socio-economic losses, and secondary disaster risks.

[0025] Specifically, the prior probabilities of the aforementioned nodes are generated based on historical data and expert experience: the historical frequency of different states of each node is statistically analyzed as the basic prior; the historical data is corrected by domain experts, and rare but important events are given reasonable weights; for new nodes or new states lacking historical data, uniform distribution or no-information priors are used as initial values, and in addition to probability statistics, initial settings can be made through relevant physical models or data deep learning methods.

[0026] The network structure is constructed using a hybrid learning framework based on conditional independence testing and score search. A complete set of nodes is defined. ,in As a disaster-causing node, As an intermediate event node, For disaster outcome nodes, firstly, based on historical event data, an improved PC algorithm is used to perform conditional independence testing. For any node pair... Given a set of conditions Below, by calculating conditional mutual information The system then uses statistical significance tests to determine if direct dependencies exist. If there is significance between nodes, a dependency exists, indicating a corresponding edge. This process starts with a completely undirected graph and gradually removes edges deemed conditionally independent, forming the network skeleton. After obtaining the network skeleton, the direction of the edges is determined through a two-layer mechanism. The first layer is a temporal constraint mechanism, establishing hard constraints based on the physical temporal logic of events: for any node... (Causing factors) and (Intermediate or result events) are only allowed to exist. The edge of the direction; for and Only allowed The second layer is a directional optimization based on a scoring function. Within a set of directed acyclic graphs satisfying temporal constraints, the optimal structure of the Bayesian network is found using the Bayesian Information Criterion (BIC) as the scoring function combined with a heuristic algorithm. The estimation of the conditional probability parameters normalizes the conditional mutual information to [0, 1].

[0027] 3. Evidence Input and Status Update Real-time monitoring data is input as evidence into the Bayesian network model to update the state of relevant nodes.

[0028] Specifically, real-time monitoring data needs to undergo necessary preprocessing and transformation before being input into a Bayesian network. This is because the raw data collected by the monitoring system is usually continuous numerical values ​​or discrete state signals, while Bayesian network nodes require probability distribution forms. This transformation process is the first step in data injection and is fundamental to ensuring the accuracy of subsequent inference.

[0029] Monitoring data is divided into two main categories: continuous and discrete. Continuous data includes physical quantities such as hourly rainfall, wind speed, temperature, and water depth, which are characterized by continuous change. Discrete data includes circuit breaker status, fault indicators, and equipment operating modes, which typically have only a limited number of possible states.

[0030] For processing continuous monitoring data, a threshold segmentation method is used. Each Bayesian network node predefines several discrete states; for example, the rainfall intensity node might define four states: "light rain," "moderate rain," "heavy rain," and "torrential rain." Each state corresponds to a numerical range; for example, "heavy rain" corresponds to rainfall of 25-50 millimeters per hour. When the monitoring data falls within a certain range, that state receives a high probability; when the data is near the boundary between two ranges, we need to assign probabilities between the two adjacent states.

[0031] Boundary probability assignment formula:

[0032] in For monitoring values, and For state The upper and lower bounds. This formula ensures the smoothness of the probability assignment, when Right at the border When above, state The probability is 1, state The probability is 0; when from Towards During movement, state The probability decreases linearly from 1 to 0, state The probability increases linearly from 0 to 1.

[0033] This boundary handling is crucial in practical applications. For example, when rainfall is 23 mm / hour, it falls precisely on the boundary between "moderate rain" (10-25 mm) and "heavy rain" (25-50 mm). Simply categorizing it into one class would result in information loss. A better approach is to consider it partly as "moderate rain" and partly as "heavy rain," assigning probabilities based on its distance from the boundary. This approach not only better reflects reality but also makes network inference smoother and more continuous.

[0034] For discrete monitoring data, the conversion is relatively simple. For example, circuit breaker status typically only has two states: "open" and "closed." When the monitoring system reports that the circuit breaker is in the open state, we directly set the probability of the "open" state for the corresponding node to 1.0 and the probability of the "closed" state to 0. This one-to-one mapping ensures accurate information transmission and can be achieved using indicator functions. express:

[0035] in This is an indicator function; its value is 1 if the condition is true, and 0 otherwise. To monitor the status, This indicates the corresponding data.

[0036] Evidence injection mechanism: Evidence injection is the essential step of inputting the preprocessed probability distribution into a Bayesian network. Based on the degree of certainty of the evidence, two injection methods are distinguished: hard evidence injection and soft evidence injection. The choice between these two methods depends on the reliability and accuracy of the monitoring data.

[0037] Hard evidence injection is suitable for observations that are absolutely certain. Examples include substation water ingress confirmed by video surveillance or line tripping events confirmed by protection device activation signals. For this type of evidence, the observations are fully trusted, and the probability distribution of the corresponding nodes is considered. This is set to an absolutely deterministic distribution for that state. Specifically, if a node has K possible states, and it is observed to be in the k-th state, then a probability vector of length K is set, where the k-th element is 1, and all other elements are 0. This representation clearly expresses the information that "the node is definitely in this state":

[0038] Soft evidence injection is suitable for observations with a certain degree of uncertainty. In actual monitoring, much data carries measurement errors or inference uncertainties. For example, the health status of equipment inferred from sensor readings, or the rainfall intensity predicted by meteorological models. For this type of evidence, we do not give an absolutely definitive judgment, but rather a probability distribution for each state. This probability distribution reflects our degree of uncertainty about the node state; states with higher values ​​are more likely, and states with lower values ​​are less likely, but the sum of the probabilities of all states must be 1.

[0039] Once evidence injection is complete, the prior probability of the corresponding node is replaced by the evidence distribution. This node becomes the information source in the network, and its state information will propagate to other connected nodes through conditional probability relationships. Multiple evidence nodes can be injected simultaneously, and their information will interact and influence each other in the network, ultimately forming a consistent inference about the state of the entire network.

[0040] Probability propagation and update algorithm: After evidence injection, the key task is to propagate the evidence information to all relevant nodes through the network structure and update their probability distributions. This process is based on Bayes' theorem and achieves information transmission through the conditional probability relationships of the network. A message-passing mechanism is used to implement this process.

[0041] In the messaging framework, each node maintains its own belief distribution, which is the current best estimate of the node's state based on all available information. Nodes share information and update their beliefs by exchanging messages. Messages are divided into two categories: π messages, which flow from parent nodes to child nodes and convey information about causality; and λ messages, which flow from child nodes to parent nodes and convey information about diagnostics.

[0042] When updating its own beliefs, each node needs to integrate prior information from its parent node and likelihood information from its child nodes. Specifically, the node uses its own conditional probability table as a transformation function, converting the parent node's state information into a prediction of its own state, while simultaneously converting the observation information from its child nodes into constraints on its own state. These two pieces of information are fused using Bayes' theorem to form the updated belief distribution. :

[0043] in As a normalization constant, ensure , Represents variable nodes Corresponding to a certain state .

[0044] Messages from parent node to child node The calculation is based on the law of total probability:

[0045] here To represent the combination of parent node states, sum and traverse all possible combinations. express Corresponding parent node, Is the parent node in State to child node The cause and effect of the message.

[0046] Messages from child node to parent node The calculation takes into account the influence of all child nodes:

[0047] in, Indicates the node index number. Indicates the first 1 node Indicates the first 1 node This represents the node state value, considering only the parent node among all possible combinations of parent node states. Just in the right state Those combinations, This represents a specific combination of states of all parent nodes. This indicates that in this combination, the parent node The state. Indicates a request The state must be Function: In summation calculations, only those... Calculate the case where ignoring This refers to other states.

[0048] Message passing is an iterative process. Initially, only evidence nodes have information, and they send messages to their neighbors. Nodes receiving messages update their beliefs and then, based on their new beliefs and the network structure, send new messages to other neighbors. Messages propagate along the edges of the network, from one node to another, gradually spreading throughout the entire network. Each node may receive messages from multiple neighbors, and it needs to rationally integrate these messages.

[0049] Typically, synchronous or asynchronous update strategies are employed. In synchronous updates, all nodes calculate the new belief simultaneously and then send the new message concurrently; in asynchronous updates, nodes update one by one, sending the new message immediately after completion. Each strategy has its advantages and disadvantages: synchronous updates are easier to implement but may converge slower, while asynchronous updates may converge faster but are more complex to implement. In moderately sized networks like urban power outage disaster chains, synchronous updates are usually sufficient.

[0050] Convergence judgment and iterative control Since message passing is an iterative process, a reasonable convergence judgment mechanism and iteration control strategy need to be designed to ensure that the algorithm can obtain stable and reliable results within a finite time. Inappropriate convergence judgment may lead to premature stopping and inaccurate results, or premature stopping and wasted computational resources.

[0051] The core idea of ​​convergence judgment is to monitor changes in the network state. It records the belief distribution of all nodes after each iteration and calculates the change between two adjacent iterations. If this change is less than a pre-set threshold, the network is considered to have reached a stable state, and iteration can stop. Commonly used metrics for change include maximum component difference and Euclidean distance; the former focuses on the maximum change of a single node, while the latter considers the overall change of all nodes.

[0052] Convergence criterion formula:

[0053] in The preset convergence threshold, The maximum component difference or Euclidean distance can be used.

[0054] In practical applications, it is necessary to set a reasonable convergence threshold. A threshold that is too small will lead to unnecessary iterations and increase the computational burden; a threshold that is too large will result in inaccurate results and affect the quality of inference. For applications such as urban power outage risk assessment, the convergence threshold is typically set at... arrive The range between these values ​​is a reasonable one, achieving a good balance between accuracy and efficiency.

[0055] In addition to the convergence threshold, a maximum number of iterations needs to be set as a safety guarantee. Some network structures may lead to slow message passing convergence, or even oscillatory non-convergence in certain situations. The maximum number of iterations ensures that the algorithm will terminate within a finite time even in the worst case. For typical catastrophe chain networks, setting a maximum number of iterations of 100-200 is usually sufficient. After these iterations, the final belief is the final posterior probability of the node's state for each node given evidence from the observed data.

[0056] 4. Probabilistic Reasoning and Risk Calculation Based on the updated Bayesian network model, conditional probability inference is performed to calculate the probability of the occurrence of power outage events and related impact events.

[0057] All probability calculations are based on the assumption that the Bayesian network nodes have completed belief updates and generated the final posterior probabilities. Let the set of network nodes be denoted as . After inputting all real-time monitoring evidence After that, each node Its posterior probability distribution has been obtained using a message passing algorithm. ,Right now The task at this stage is to systematically calculate the probability of occurrence of various power outage events and related impact events that users are concerned about, based on these posterior distributions.

[0058] Target event Defined as network state space A subset, for example This indicates that "a power outage has occurred." All reasoning is transformed into considering posterior probabilities. The calculation. For simple events, such as "node". In state Its posterior probability can be directly read as For continuous risk indicator nodes, the event "risk value exceeds the threshold" is considered a risk indicator node. The probability of "" is calculated as follows:

[0059] That is, summing the posterior probabilities of all discrete states that exceed the threshold.

[0060] A power outage event may be directly mapped to a single node, or it may be the logical result of a series of preceding events. For direct mappings, such as the "Power Outage in Area A" node... Its posterior probability This is what we are looking for. For cascading failures, define a set of critical intermediate events. Then the power outage event It can be represented as A Boolean function for an event Its posterior probability is:

[0061] in It is an indicator function, summing and traversing. All combinations of states , This represents the joint probability calculated based on the current posterior distribution. To reduce computational complexity, it is typically summed only for key state combinations with significantly non-zero probabilities.

[0062] Secondary impacts of power outages (Such as traffic disruptions and communication failures) typically depend on the state of the power system. Its posterior probability The calculation is performed using the formula of total probability and the obtained posterior probability of the power event:

[0063] in These are conditional probabilities (such as CPT) defined in the network, representing the probability under a given power state. Downward-impacting events The probability of occurrence. It is the joint posterior probability of the power state, derived from network inference. For nodes with comprehensive socioeconomic impacts, they are themselves nodes in the network, and their posterior probabilities... The posterior probability of the belief update is given directly.

[0064] 5. Risk Assessment and Results Output Based on the reasoning results, the power outage risk is classified and assessed, and information on the risk level and disaster chain evolution path is output.

[0065] Based on the above content, the posterior probabilities of various events obtained through Bayesian network inference are... The system establishes a quantitative comprehensive risk index. This is then mapped to discrete risk levels. The comprehensive risk index calculation considers the probability of the event, the degree of impact, and the depth of the disaster chain propagation:

[0066] in For the event The weight reflects its importance and satisfies The propagation penalty coefficient (usually) ); To normalize the average propagation depth of the disaster chain, The average length of the critical path in the current posterior state. This represents the maximum possible path length in the network.

[0067] Risk level Determined by threshold division:

[0068] Regional corrections introduce vulnerability coefficients. and importance coefficient Implementation, region The final risk index is ,in , and These are the regional vulnerability index and the importance index, respectively. and This is the citywide average.

[0069] Evolutionary path Defined as a node sequence Its probability is determined by the conditional probability of the states of the nodes on the path and the current posterior state:

[0070] The above is the calculation based on the given evidence. Under these conditions, the evolution path of the disaster chain The formula for the probability of occurrence: Path starting point posterior probability Multiply by each subsequent node on the path Given its predecessor node and all its other parent nodes Conditional posterior probability in a state The product of consecutive products.

[0071] To find the most probable path, the system employs a two-stage dynamic programming algorithm. The first stage (forward computation) involves calculating the path for each node. and path length Calculate the maximum score:

[0072] in for The set of direct predecessor nodes is used to select the Top-K paths with the highest scores as disaster chain evolution path information.

[0073] The system solution provided by this invention: To implement the above method, the present invention also provides a corresponding system structure, including but not limited to: The system includes: a data access module; a disaster chain modeling module; a probabilistic reasoning module; a risk assessment module; and a results output module. These modules can be centrally deployed on the emergency command center server, or they can be deployed in collaboration between edge nodes and the central platform.

[0074] To realize the aforementioned method for dynamic reasoning and risk assessment of urban large-scale power outage disaster chains based on Bayesian networks, this invention provides a corresponding system implementation scheme. The system adopts a modular design, mainly including a data access module, a disaster chain modeling module, a probabilistic reasoning module, a risk assessment module, and a result output module. These modules work closely together logically, forming a complete processing loop from multi-source data input to risk assessment output. The data access module is responsible for real-time collection and aggregation of multi-source heterogeneous data from meteorological monitoring stations, hydrological sensors, power grid SCADA systems, and urban operation platforms, and performs preprocessing such as spatiotemporal alignment, missing data filling, and anomaly cleaning. The disaster chain modeling module constructs and maintains a Bayesian network model representing the causal relationship between "disaster-causing factors-intermediate events-disaster consequences" based on historical cases and expert knowledge, and supports dynamic learning and updating of network structure and parameters. The probabilistic reasoning module receives real-time monitoring data as evidence, dynamically updates the network state and performs probability propagation based on connection tree algorithms or approximate reasoning methods, and calculates the posterior probability of various power outages and related events. The risk assessment module calculates the regional comprehensive risk index and performs graded assessment based on the probabilistic reasoning results, combined with weighting factors such as spatial vulnerability and socio-economic benefits, and extracts the evolution path and temporal characteristics of key disaster chains. The results output module generates structured reports, visual charts, and early warning instructions for risk levels, evolution paths, key causes, and disposal recommendations, and pushes them to emergency command systems, mobile terminals, and related business platforms through API interfaces, message queues, and other means. In terms of deployment, all modules can be centrally deployed on the emergency command center server, or they can be deployed at the edge. The "central" collaborative architecture deploys lightweight data access and real-time inference units at edge nodes such as power grid substations and drainage pumping stations to enable rapid local analysis and response; and deploys complete modeling, evaluation, and output modules on the central platform to conduct global collaborative analysis and decision support, thereby balancing real-time performance, reliability, and system scalability.

[0075] This invention is applicable to, but not limited to, the following scenarios: 1. Urban power emergency protection and risk assessment; 2. Power outage risk prediction under extreme weather conditions; 3. Urban emergency command and dispatch decision support; 4. Urban comprehensive disaster prevention and mitigation information system.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for urban disaster reasoning and risk assessment based on Bayesian networks, characterized in that, include: Acquire real-time monitoring data from multiple sources, including urban infrastructure and meteorological data; Acquire historical event data and construct a Bayesian network based on the historical event data, wherein the Bayesian network includes disaster-causing factor nodes, intermediate event nodes, and disaster result nodes; The real-time monitoring data is used as evidence to input into the Bayesian network model, and the state of the relevant nodes in the Bayesian network is updated. Conditional probability inference is performed based on the updated Bayesian network to calculate the probability of occurrence of power outage events and related impact events; The risk of power outages is classified and assessed based on the probability of occurrence, resulting in information on the risk level and the corresponding disaster chain evolution path.

2. The method according to claim 1, characterized in that, The real-time monitoring data includes power grid operation data, meteorological data, hydrological data, and urban infrastructure operation status data.

3. The method according to claim 1, characterized in that, The process of constructing a Bayesian network includes: Based on historical event data, corresponding disaster-causing factor nodes, intermediate event nodes, and disaster outcome nodes are assigned. Conditional independence is tested using conditional mutual information based on historical event data, and a network skeleton is formed for different nodes. The network skeleton is optimized based on temporal constraint mechanism and scoring function optimization, and prior probabilities are set for different nodes to generate a Bayesian network.

4. The method according to claim 1, characterized in that, The process of inputting the real-time monitoring data as evidence into the Bayesian network model includes: The real-time monitoring data is transformed to generate corresponding state probabilities. Hard evidence is injected into the direct observation results in the real-time monitoring data, and soft evidence is injected into the uncertain data in the real-time monitoring data to generate corresponding probability distribution evidence. After injection, the prior probabilities of the nodes of the corresponding Bayesian network are replaced with probability distribution evidence.

5. The method according to claim 1, characterized in that, The process of updating the state of the relevant nodes in the Bayesian network includes: Based on the input evidence, the belief distribution of all nodes in the Bayesian network is updated using a message passing algorithm. The message passing algorithm includes: passing causal information from parent nodes to child nodes along directed edges of the network via a first message, and passing diagnostic information from child nodes to parent nodes along directed edges of the network via a second message. Each node calculates its updated belief distribution as the posterior probability of its node state in the Bayesian network based on the received first and second messages and its own conditional probability table.

6. The method according to claim 1, characterized in that, The calculation process for the probability of power outage events and related impact events includes: Based on the posterior probability distribution of each node in the updated Bayesian network, the probability of a power outage event is calculated by probability summation and conditional probability propagation. Specifically, for a power outage event defined as a single node in a specific state, its probability of occurrence is directly read from the posterior probability of the corresponding state of that node; for a cascading power outage event defined as a logical combination of the states of several nodes, its probability of occurrence is obtained by summing the posterior joint probabilities of all related state combinations of the several nodes. For secondary impact events of a power outage, the probability of occurrence of the secondary impact event is calculated using the total probability formula, which combines the posterior probability of the power system state and the conditional probability of the secondary impact event under a given power system state.

7. The method according to claim 1, characterized in that, The process of classifying and assessing power outage risks based on their probability of occurrence includes: The comprehensive risk index is calculated based on the probability of occurrence and the corresponding depth of the disaster chain propagation. The comprehensive risk index is then classified into different thresholds to obtain the risk level.

8. The method according to claim 1, characterized in that, The process of obtaining the disaster chain evolution path information includes: The propagation path of the Bayesian network nodes corresponding to the power outage event is evaluated, and the corresponding path score is generated. Based on the path score, the corresponding disaster chain evolution path information is selected.

9. A Bayesian network-based urban disaster reasoning and risk assessment system, characterized in that, Used to perform the method described in any one of claims 1-8.

Images