Torrential rain induced highway network inspection and recovery joint decision method, device, medium and product

Abstract

The application discloses a heavy rain disaster-causing highway network inspection and recovery joint decision method, equipment, medium and product, relates to the traffic infrastructure resilience recovery and intelligent decision-making technical field. The method constructs a fixed rough decision graph of the recovery section; a candidate inspection shortlist is obtained through pre-screening; an activated rough decision subgraph containing the candidate inspection section is extracted from the fixed rough decision graph, the local posterior offset is propagated to the neighborhood within the activated rough decision subgraph according to the bounded support kernel, and an updated rough space posterior state is obtained; after being converted into an abstract state vector, the updated rough space posterior state is input into a short-time recovery value reduction evaluator to obtain an updated short-time recovery value, so that a short-time recovery information value is obtained; the candidate inspection section with the maximum short-time recovery information value is selected to perform inspection, the repair priority is updated based on the inspection result, and a rolling repair decision is performed. The application can improve the key accessibility recovery efficiency and reduce the online calculation overhead under the conditions of incomplete information and limited resources.

Description

Technical Field

[0001] This application relates to the field of transportation infrastructure resilience restoration and intelligent decision-making technology, and in particular to a method, equipment, medium and product for joint decision-making on highway network inspection and restoration caused by rainstorm disasters. Background Technology

[0002] Under the influence of torrential rains and their secondary disasters, highway networks are prone to coupled or cascading failures of multiple disasters, such as floods, washouts, landslides, collapses, debris flows, and subsidence, causing a rapid decline in road capacity or even complete interruption. In the early stages of a disaster, management departments typically face constraints such as incomplete road condition information, limited on-site inspection capabilities, insufficient coverage of fixed monitoring equipment, rapid evolution of road conditions, and a shortage of maintenance resources. Therefore, they must coordinate and answer three interconnected questions within a limited time: where to inspect first, where to repair first, and how to allocate maintenance resources.

[0003] Existing technologies typically model road disaster state inference, inspection target selection, and repair scheduling separately: one type of method focuses on using machine learning, graphical models, or statistical models to predict whether roads will be interrupted; another type focuses on post-disaster repair prioritization and resource scheduling; and yet another type attempts to incorporate observational or informational value, but most of these methods revolve around improving the accuracy of state recognition and have not yet directly anchored the value of observational actions to short-term post-disaster recovery benefits. This leads to an inconsistency between the objectives of inspection and repair decisions, making it difficult to ensure that the limited inspection budget truly serves to improve recovery efficiency.

[0004] In disaster recovery scenarios, the value of candidate inspection actions depends not only on the updated state assessment of the observed road segment itself, but also on the interconnected impact of the observation results on the posterior state of the relevant neighborhood, and on how this impact alters the subsequent repair sequence, critical accessibility, and service satisfaction. Therefore, the value of inspection actions needs to be defined uniformly around the recovery objectives, rather than solely based on reduced uncertainty or improved identification accuracy.

[0005] Furthermore, under the conditions of multi-hazard cascading disasters induced by rainstorms, adjacent or functionally related road sections typically share terrain control, drainage paths, disaster corridors, and critical service functions, exhibiting significant recovery-related coupling relationships. If such coupling relationships are ignored, inspection results cannot be effectively transmitted to the neighborhood state that truly affects recovery benefits; if the complete posterior and repair benefits are recalculated at the entire road network scale for each candidate inspection action, the online computation cost is too high, making it difficult to meet the rolling solution timeliness requirements of emergency scenarios.

[0006] Therefore, there is an urgent need for a joint decision-making method and system for highway network inspection and restoration that can jointly characterize inspection selection and repair ranking under a unified restoration objective, express restoration-related coupling relationships, and approximate the benefits of candidate inspection actions at a low online cost. Summary of the Invention

[0007] The purpose of this application is to provide a method, equipment, medium, and product for joint decision-making on highway network inspection and restoration caused by rainstorms, which can improve the efficiency of critical accessibility restoration and reduce online computing overhead under conditions of incomplete information and limited resources.

[0008] To achieve this objective, this application provides the following solution: Firstly, this application provides a joint decision-making method for highway network inspection and restoration in response to rainstorm disasters, including: Obtain basic data on the highway network to be restored; Construct a fixed coarse decision map of the road sections to be restored in the road network to be restored; Based on the basic data, determine the posterior probability that each restored road segment belongs to a different service state; Based on the posterior probability, the restored road sections are pre-screened to obtain a short list of candidate inspections; Based on the observation model and the observation results of different potential inspections, the local posterior offset of each candidate inspection segment in the shortlist of candidate inspections is determined. Extract the activated coarse decision subgraph containing candidate inspection segments from the fixed coarse decision graph, construct a bounded support kernel using the adjacency matrix of the activated coarse decision subgraph, and propagate the local posterior offset to the neighborhood within the activated coarse decision subgraph based on the bounded support kernel to obtain the updated coarse spatial posterior state. After converting the updated coarse space posterior state into a summary state vector, it is input into the short-time recovery value down-order evaluator to obtain the updated short-time recovery value. The expected value of the difference between the updated short-term recovery value and the baseline short-term recovery value before observation is obtained to obtain the short-term recovery information value of each candidate inspection segment. The candidate road sections with the highest short-term recovery information value are selected from the shortlist of candidate inspections and inspected. The repair priority of each object to be repaired is updated based on the inspection results, and rolling repair decisions are made according to the updated repair priority under the constraints of maintenance resources.

[0009] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the above-mentioned joint decision-making method for highway network inspection and restoration caused by rainstorm disasters.

[0010] Thirdly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned joint decision-making method for highway network inspection and restoration caused by rainstorm disasters.

[0011] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned joint decision-making method for highway network inspection and restoration caused by rainstorm disasters.

[0012] This application has the following technical effects: This application defines the value of candidate inspection actions as their expected marginal improvement in short-term recovery benefits, allowing inspection target selection to directly revolve around the recovery objective. By constructing a fixed coarse decision graph offline, road segments related to recovery are compressed into a smaller decision space, and only local coarse decision subgraphs are activated during the online phase, balancing the expressive power of recovery relevance with online computational efficiency. Through bounded-supported graph coupling of posterior transmission, local observation results are transformed into posterior linkage updates of recovery-related neighborhoods, thereby more accurately assessing the impact of inspection actions on subsequent recovery decisions. By replacing the full-network recalculation of each candidate inspection action with a short-term recovery value downgrade evaluator, and unifying inspection decisions and rolling repair decisions under the same recovery benefit objective, critical accessibility and service capabilities can be restored more quickly under limited inspection budgets and maintenance resources. Attached Figure Description

[0013] Figure 1 A simplified flowchart illustrating a joint decision-making method for highway network inspection and restoration in response to rainstorm disasters; Figure 2 A schematic diagram illustrating the process of constructing a fixed coarse decision diagram; Figure 3 This is a schematic diagram of the graph-coupled posterior transport update process; Figure 4 This is a schematic diagram of the structure of a computer device. Detailed Implementation

[0014] To address the issues of disconnect between the value definition of inspection actions and restoration targets, insufficient utilization of restoration-related coupling relationships, and high computational costs for online joint decision-making in existing technologies, an exemplary embodiment provides a joint decision-making method for highway network inspection and restoration in response to rainstorm disasters. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method includes steps 101 to 109.

[0015] Step 101: Obtain basic data of the highway network to be restored.

[0016] Step 102: Construct a fixed coarse decision map of the road segments to be restored in the road network to be restored.

[0017] Step 103: Based on the basic data, determine the posterior probability that each restored road segment belongs to a different service state.

[0018] Step 104: Based on the posterior probability, pre-screen the restored road sections to obtain a short list of candidate inspections.

[0019] Step 105: Based on the observation model and the observation results of different potential inspections, determine the local posterior offset of each candidate inspection segment in the shortlist of candidate inspections.

[0020] Step 106: Extract the activated coarse decision subgraph containing candidate inspection segments from the fixed coarse decision graph, construct a bounded support kernel using the adjacency matrix of the activated coarse decision subgraph, and propagate the local posterior offset to the neighborhood within the activated coarse decision subgraph based on the bounded support kernel to obtain the updated coarse spatial posterior state.

[0021] Step 107: After converting the updated coarse space posterior state into a summary state vector, input it into the short-term recovery value downgrade evaluator to obtain the updated short-term recovery value.

[0022] Step 108: Calculate the expectation of the difference between the updated short-term recovery value and the baseline short-term recovery value before observation to obtain the short-term recovery information value of each candidate inspection segment.

[0023] Step 109: Select the candidate road segment with the highest short-term recovery information value from the short list of candidate inspections and perform inspections. Update the repair priority of each object to be repaired based on the inspection results, and execute rolling repair decisions according to the updated repair priority under the constraints of maintenance resources.

[0024] By implementing steps 101 to 109 above, this application unifies the selection of inspection actions and the sequencing of repair actions with the short-term recovery benefit objective, which can improve the efficiency of key accessibility recovery and reduce online computing overhead under conditions of incomplete information and limited resources.

[0025] In another exemplary embodiment of this application, the basic data in step 101 includes: highway network data, historical disaster point data, hydrological and meteorological forcing data, road condition confirmation data, restoration resource data, and critical service demand data. Highway network data includes at least one or more of the following: route number, station number, road grade, number of lanes, road width, bridge or culvert markers, design speed, landform type, and high slope attributes. Historical disaster point data includes at least one or more of the following disaster types: landslide, erosion, subsidence, debris flow, and landslide; disaster location; impact range; historical handling results; and interruption records. Restoration resource data is used to characterize the available post-disaster maintenance forces and their operational capabilities, including at least the number of maintenance teams, team type, type and quantity of machinery and equipment, material reserves, resource occupancy per unit of repair object, estimated repair time, construction accessibility constraints, and resource scheduling time window. The restoration resource data is subsequently used to determine the feasibility of candidate repair actions, calculate resource occupancy and estimated repair time, and serves as a maintenance resource constraint in rolling repair decisions. Key service demand data should include at least one or more of the following: key origin-destination pairs, emergency facility locations, traffic control corridors, and service priority information.

[0026] The basic data undergoes coordinate unification, time alignment, missing data repair, and structured coding to form standardized inputs for subsequent offline modeling, feasibility assessment of candidate repair actions, online decision-making, and online rolling repair decision-making.

[0027] In another exemplary embodiment of this application, such as Figure 2 As shown, step 102 can be replaced by the following steps 201 to 205.

[0028] Step 201: Construct a directed road segment map based on the road network data in the basic data.

[0029] Step 202: Select a subset of restored road segments from the directed road segment map; the subset of restored road segments must meet the following conditions: located on critical origin-end point paths or emergency access routes, intersecting with historical high-risk disaster corridors, serving as one of the alternative routes to bypass historical high-risk disaster corridors, or having an impact on the satisfaction of critical service needs that is greater than the impact threshold; and the risk disaster incidence rate of the high-risk disaster corridor is greater than the risk threshold.

[0030] Step 203: Determine the service correlation and disaster correlation of road segment pairs in the restored road segment subset, and merge the service correlation and disaster correlation to obtain static pairwise similarity.

[0031] The formula for calculating static pairwise similarity is: ; Let be the static pairwise similarity between road segment i and road segment j. To improve service relevance, For disaster correlation, The fusion weight between service correlation and disaster correlation, and 0 ≤ α ≤1.

[0032] A road segment pair consists of two different road segments from a subset of recovered road segments. Two road segments are identified as a single road segment pair when they are located on the same critical origin-destination path, the same emergency access route, the same disaster corridor, the same route, or adjacent routes, or when the chain distance or geographical distance between the two road segments is less than a preset threshold. This ensures both service relevance and disaster relevance while reducing offline computation.

[0033] Service relevance characterizes the combined impact of road segments on the fulfillment of critical accessibility or critical service demands. The process of determining service relevance is as follows: obtain the set of origin-destination paths, the set of facilities served, and the set of corridors affected by the two road segments in the road segment pair; calculate the overlap of the origin-destination path sets, the facility sets, and the corridor sets for the two road segments respectively; and perform a weighted summation of the overlap of the origin-destination path sets, the facility sets, and the corridor sets to obtain the service relevance.

[0034] Disaster correlation characterizes the common disaster tendency of road segments in terms of disaster occurrence mechanism, spatial proximity, drainage or topographic unit. The process of determining disaster correlation is as follows: calculate the consistency of dominant disaster type, spatial distance attenuation term, drainage unit consistency and route category consistency of the two road segments in the road segment pair respectively; after normalizing the consistency of dominant disaster type, spatial distance attenuation term, drainage unit consistency and route category consistency, the weighted sum is obtained to obtain disaster correlation.

[0035] Step 204: While maintaining the connectivity constraints of the directed road segment graph, perform agglomerative clustering on the recovered road segment subset based on static pairwise similarity to obtain multiple coarse nodes.

[0036] Step 205: Construct coarse edges based on the original adjacency relationship of the restored road segments within the coarse nodes or when the average similarity between coarse nodes is greater than the preset similarity threshold, thus forming a fixed coarse decision graph.

[0037] The average similarity between any two coarse nodes is obtained by averaging the static pairwise similarities between the member road segments within the two coarse nodes. For any two coarse nodes, if their member road segments are adjacent or connected in the original directed road segment graph, or if their average similarity is greater than a preset similarity threshold, then a coarse edge is established between them. The weight of the coarse edge is determined by the original adjacency strength of the member road segments, the average similarity between coarse nodes, or a weighted combination of the two.

[0038] After forming a fixed coarse decision graph, the mapping relationship between coarse nodes and member road segments, as well as the neighborhood search structure, are recorded. The coarse node-member road segment mapping relationship records the correspondence between each coarse node and its constituent original directed road segment set; each restored relevant road segment uniquely belongs to a coarse node. This mapping relationship supports the projection of road segment-level observation results onto coarse nodes, and the back-projection of coarse spatial posterior states or repair priorities onto specific road segments. The neighborhood search structure includes the adjacency list, coarse edge weight table, shortest hop count table between coarse nodes, and neighborhood index table within a preset support depth of the fixed coarse decision graph. This is used to quickly determine the support depth neighborhood of the coarse node to which a candidate inspection road segment belongs during the online phase. For any coarse node... C q The neighborhood search structure can quickly return its first-order neighborhood, second-order neighborhood, and even the set of coarse nodes within a preset support depth d. N d ( C q This is used to quickly extract the coarse decision subgraph for activation during the online phase.

[0039] In another exemplary embodiment of this application, step 103 may be replaced by steps 301 to 303.

[0040] Step 301: Construct a priori disaster conditions under different service states based on the highway network data, historical disaster point data, and key service demand data in the basic data; the priori disaster conditions include the prior risk values ​​of each restored road section.

[0041] Constructing a priori disaster conditions involves: projecting historical disaster point data onto corresponding road segments in a directed road segment map according to spatial location, route number, and station number; extracting historical disaster frequency, disaster type, impact range, interruption records, landform type, high slope attributes, bridge or culvert markers, drainage units, rainfall sensitivity, and critical service demand exposure for each road segment; normalizing and encoding the features; and using rule-weighted, statistical, or machine learning models to calculate the prior risk values ​​of each road segment under different service states, thus forming a road segment-level priori disaster conditions.

[0042] Step 302: Based on the highway network data, hydrological and meteorological forcing data, road condition confirmation data and historical disaster point data in the basic data, use the posterior inference model to calculate the posterior logarithmic values ​​of each restored road segment under different service states; different service states include traffic status, downgraded traffic status and closed status.

[0043] The posterior inference model can employ a spatiotemporal graph neural network model or other spatiotemporal inference models capable of incorporating temporal forcing, spatial topology, and observational evidence. The posterior inference model must incorporate at least the following four types of information: dynamic forcing information, road condition observation information, static terrain and road attribute information, and historical disaster evidence information; among which, dynamic forcing information includes one or more of the following: multi-time-window rainfall intensity, cumulative rainfall, and runoff proxy.

[0044] Step 303: Superimpose the posterior logarithmic values ​​of different service states with the prior field of disaster conditions for each service state and normalize them to obtain the posterior probability of each restored road segment belonging to different service states.

[0045] The formula for calculating the tri-state posterior probability is: ; Let be the posterior probability that road segment i is in state s. This represents the logarithmic state output of the posterior inference model. These are the corresponding state priors given by the disaster condition prior field. This is a set of preset service states.

[0046] In another exemplary embodiment of this application, step 104 may be replaced by steps 401 to 405.

[0047] Step 401: Use the posterior probability of the closed state as the closed state probability.

[0048] Step 402: Determine the posterior entropy based on the posterior probability.

[0049] Posterior entropy is used to characterize state uncertainty.

[0050] Step 403: Determine the time interval or normalized result of the time interval between the current time and the most recent valid road status confirmation time of the restored road segment, and use it as the observation time limit.

[0051] Step 404: Weight the closed-state probability term, posterior entropy, service criticality, and observation timeliness to obtain the pre-screening score for each restored road segment. Service criticality describes the importance of a single road segment to critical accessibility, emergency services, and critical corridors.

[0052] The formula for calculating the pre-screening score is: ; Candidate inspection sections Pre-screening scores This represents the posterior probability that the candidate inspection section is in a closed state. For its posterior entropy, Its service is critical. Its observation timeframe or the length of time since the most recent valid confirmation, to The non-negative weights are determined offline.

[0053] Step 405: Select road sections with pre-screened scores not less than the score threshold according to the preset screening rules, or select no more than a preset number of road sections according to the pre-screened scores from high to low to form a short list of candidate inspection sections.

[0054] In another exemplary embodiment of this application, step 105 may be replaced by steps 501 to 502.

[0055] Step 501: Based on the conditional probability between the reported state and the potential true state corresponding to the inspection results in different potential inspection observation results, the inspection observation model is used to determine the local posterior logarithmic value after observation.

[0056] The inspection observation model characterizes the conditional probability relationship between the inspection report status and the potential true status. The inspection observation model adopts the inspection confusion matrix model, which, given the true status as passable, downgraded passable, or closed, provides the conditional probability of the inspection result being reported as a passable report, downgraded passable report, closed report, or indeterminate report, respectively.

[0057] Step 502: Calculate the difference between the local posterior logarithmic value after observation and the local posterior logarithmic value before observation to obtain the local posterior offset of each candidate inspection segment in the short list of candidate inspection segments.

[0058] In another exemplary embodiment of this application, step 106 extracts an activated coarse decision subgraph from the constructed fixed coarse decision graph, containing the coarse nodes to which the candidate inspection road segment belongs and their preset support depth neighborhoods (the activated coarse decision subgraph is locally activated based on the fixed coarse decision graph formed offline to reduce online computational overhead); a bounded support kernel is constructed using the normalized adjacency matrix of the activated coarse decision subgraph, and the local posterior offset is propagated to the recovery-related neighborhood within the activated coarse decision subgraph according to the bounded support kernel to obtain the updated coarse spatial posterior state. The adjacency matrix is ​​the coarse node connection matrix corresponding to the activated coarse decision subgraph. When there is a coarse edge between two coarse nodes, the corresponding element in the adjacency matrix takes the weight of that coarse edge or takes 1; otherwise, it takes 0. The calculation formula for the bounded support kernel is: ; To support a bounded support core of depth d, The attenuation coefficient corresponding to the r-th order adjacent term. To activate the normalized adjacency matrix of the coarse decision subgraph, It represents the adjacency order.

[0059] like Figure 3 As shown, the coarse space posterior state update process can be replaced by the following steps 601 to 604.

[0060] Step 601: Project the local posterior offset onto the corresponding coarse node in the activated coarse decision subgraph.

[0061] Step 602: Extract the kernel vector corresponding to the coarse node in the bounded support kernel, and normalize the kernel vector in the neighborhood of the preset support depth to obtain the normalized weight.

[0062] Step 603: Multiply the local posterior offset by the normalized weight of the corresponding neighboring coarse nodes to obtain the posterior logarithmic increment of each neighboring coarse node.

[0063] Step 604: The posterior logarithmic increment is superimposed on the local posterior logarithmic values ​​of each neighborhood coarse node before observation, and normalized to obtain the updated coarse spatial posterior state of each coarse node.

[0064] The formulas for local posterior offset and neighborhood update are as follows: ; Let y be the posterior logarithmic increment of the coarse neighbor node j under the potential observation y. The local posterior offset of the source coarse node c to which the candidate inspection segment belongs. The normalized propagation weights are derived from the bounded supporting kernel. Let c be the set of neighboring coarse nodes of the source coarse node c within the support depth d.

[0065] In another exemplary embodiment of this application, the updated coarse spatial posterior states under different potential observation results are respectively converted into low-dimensional summary state vectors. The low-dimensional summary state vectors include: the closure quality value and degradation quality value of the top k coarse nodes with the highest probability of activating the closed state and the degraded access state in the coarse decision subgraph; the current key accessibility index; the current proportion of served demand; the remaining maintenance resource capacity; the average posterior entropy of activating the coarse decision subgraph; the dispersion of the posterior entropy of activating the coarse decision subgraph; and one or more of the average closure quality of activating the coarse decision subgraph.

[0066] In another exemplary embodiment of this application, when calculating the short-term recovery information value, the expected value of the difference between the updated short-term recovery value and the baseline short-term recovery value before observation is calculated using the probability of occurrence of different potential observation results (the probability of occurrence is calculated based on the current three-state posterior probability of the candidate inspection segment and the inspection observation model) as weights, and this expectation is used as the short-term recovery information value of the corresponding candidate inspection segment. ; Candidate inspection sections The value of short-term recovery information Y For the set of potential observations, To calculate the probability of obtaining a potential observation y after inspecting the candidate inspection section. To update the state based on the observation result y The corresponding short-term recovery value, Pre-observation baseline state Its short-term recovery value.

[0067] The short-term recovery value de-ranking evaluator is a machine learning model trained using offline short-term rolling simulation samples. The offline short-term rolling simulation samples are generated under multiple posterior scenarios using a fixed fast repair strategy, and each simulation sample includes at least one set of low-dimensional summary states and corresponding short-term recovery value labels. The short-term recovery value labels represent one or more of the following within a preset rolling window: the amount of key accessibility recovery, the amount of service demand fulfillment, or resource utilization efficiency.

[0068] The short-term recovery value reduction evaluator is a multilayer perceptron model, but other supervised learning models can also be used. It is trained using mean squared error loss, combined with an early stopping mechanism and regularization constraints.

[0069] In another exemplary embodiment of this application, the target road segment with the highest short-term recovery information value is selected from the shortlist of candidate inspections for inspection. The repair priority of each object to be repaired is updated based on the inspection results, and a rolling repair decision is executed under given maintenance resource constraints. The rolling repair decision determines the execution order of candidate repair actions through repair priority. Objects to be repaired include one or more of the following: closed road segments or degraded traffic sections that are in a closed or downgraded state and have repair feasibility; degraded traffic sections; washed-out roadbeds; damaged pavements; damaged bridges; damaged culverts; blocked drainage facilities; landslides; mudslide deposits; subsidence areas; flooded road sections; and traffic control points requiring temporary traffic control measures.

[0070] The repair priority is calculated by weighting the predicted improvement in critical accessibility from the repair action, the current severity of the outage to be repaired, and the estimated repair duration; or by weighting the predicted improvement in critical accessibility from the repair action, the degree of service loss to the to be repaired, and the estimated repair duration. The predicted improvement in critical accessibility and the current severity of outage or service loss are used to increase the repair priority, while the estimated repair duration is used to decrease the repair priority. The relevant weights are preset weights or weights determined offline. The formula for rolling repair priority is: ; Candidate repair actions Rolling repair priority, To predict the increase in key accessibility after performing this repair action. The current severity of the outage or the extent of service loss for the object to be repaired. To estimate the repair time, ω 1 to ω 3 represents the preset weight.

[0071] The method described in this application is used for emergency inspection and restoration of trunk roads, ordinary roads, mountain roads, or urban expressway networks under cascading conditions of multiple disasters such as floods, washouts, landslides, collapses, debris flows, and subsidence induced by rainstorms.

[0072] Figure 1 The following is a brief flowchart of the method described in this application: S1 Data Collection: Collect road network status, disaster information, inspection resources, and key requirements; S2 Offline Modeling: Construct a prior disaster scenario and form a stable coarse-grained decision structure; S3 Posterior Inference: Estimate the posterior road status and distinguish between passable, restricted, and blocked conditions; S4 Inspection Screening: Screen objects based on risk, urgency, criticality, and information uncertainty; S5 Graph Coupling Update: Combine local observation results to propagate and correct judgments in neighboring areas; S6 Value Assessment: Form a low-dimensional state summary and assess the benefits of short-term emergency repair and restoration; S7 Inspection and Repair: Determine inspection targets and continuously adjust the repair sequence and resource allocation.

[0073] The method of this application will be further explained below with an example of a regional trunk highway network.

[0074] Under the influence of continuous heavy rainfall or short-term extreme rainfall, the highway network in this region may experience various disasters such as flooding, waterlogging, landslides, collapses, mudslides, and subsidence, resulting in different service states for some road sections, including normal traffic, downgraded traffic, or closure. For ease of description, road units with unidirectionally identifiable service states are denoted as directed road segments, and all directed road segments constitute the directed road segment graph G=(V,E).

[0075] First, basic data on the regional highway network is acquired, and then cleaned, aligned, and encoded under a unified spatial reference and temporal benchmark. Hydrometeorological forcing data includes rainfall intensity, cumulative rainfall, and optional runoff proxy under different time windows; road condition confirmation data includes closure records, inspection records, UAV confirmation information, speed anomaly records, and response update information.

[0076] A directed road segment graph is constructed based on highway network data, where each directed edge corresponds to a directed road segment whose service status can be independently determined. For any directed road segment i, its potential service status set is preferably set as S = {accessible, degraded accessible, closed}; where the accessible status indicates that it can be accessed at normal capacity, the degraded accessible status indicates that it can be accessed with speed limits, load limits, one-way traffic, or controlled traffic, and the closed status indicates that it cannot provide access services.

[0077] Based on the spatial correspondence between historical disaster point data and the highway network, disaster points are projected onto corresponding road segments to obtain segment-level historical disaster evidence. Prior features of disaster conditions are formed by combining road segment topography, drainage, road grade, bridge and tunnel attributes, and high slope attributes, and a disaster condition prior field is constructed. Then, a posterior inference model is built, outputting the posterior logarithmic values ​​of the three-state service states for each road segment at a preset prediction time or within the prediction time domain. These values ​​are then superimposed with the prior field and normalized to obtain the three-state posterior probabilities. To facilitate online deployment, the posterior inference model can be trained offline, with only inference performed online.

[0078] To express the recovery of relevant coupling relationships at a lower online cost, a fixed coarse decision graph is constructed offline.

[0079] Subsequently, static pairwise similarity is calculated for road segment pairs within the restored relevant road segment subset. To reduce offline computation, service correlation and disaster correlation are calculated only for road segment combinations that meet the conditions of spatial proximity, service relevance, or disaster relevance.

[0080] Service relevance can be calculated using the following formula: ; Let i be the service correlation degree between road segment i and road segment j; , These are the sets of key origin-end paths involved in road segment i and road segment j, respectively. , Each refers to a collection of key facilities serving both. , These are the key corridor sets influenced by both; J The degree of overlap is set, preferably Jaccard overlap; , and The weights are non-negative.

[0081] Disaster correlation can be calculated using the following formula: ; The degree of disaster correlation between road segment i and road segment j; I For indicator functions; and The dominant disaster type; This refers to the distance between chain piles or geographical distance. L The spatial attenuation scale; and For drainage units or terrain units; and Group by route category or corridor; to The weights are non-negative.

[0082] Secondly, while maintaining the connectivity constraints of the original road network, agglomerative clustering is performed on the recovered relevant road segment subsets, allowing only clusters that remain connected in the original road map to be merged. Clustering stops when the number of coarse nodes reaches the target size or the minimum similarity within a cluster is below a threshold, with each cluster corresponding to one coarse node. Subsequently, coarse edges are constructed based on the original connectivity relationships between member road segments within the coarse node and the average similarity between coarse nodes, forming a fixed coarse decision graph. A mapping relationship between coarse nodes and member road segments, coarse node attribute vectors, and a neighborhood search structure are also established to quickly activate local coarse decision subgraphs during the online phase.

[0083] The average similarity between coarse nodes can be calculated using the following formula: ; coarse node With coarse nodes Average similarity between them; | | and | | These represent the number of member road segments within the two coarse nodes; Let be the static pairwise similarity between road segment i and road segment j.

[0084] During the online phase, at each rolling decision time, all inspectable candidate road segments are pre-screened based on the tri-state posterior probabilities output by the posterior inference model. For each candidate road segment, a pre-screening score is calculated.

[0085] Based on the pre-screening scores, the top k candidate road segments are selected to form a shortlist of candidate inspection segments. A closed-state probability term is used to highlight high-risk interruption segments, a posterior entropy term characterizes the uncertainty of the current state judgment, and an observation timeliness term is used to suppress decision-making bias caused by long-term unconfirmed states. Specifically, the closed-state probability is directly obtained from the three-state posterior probabilities of the candidate inspection segments; the posterior entropy is calculated based on the information entropy of the three-state posterior probabilities; the service criticality can be obtained by weighting the number of times the segment participates in critical paths, the number of emergency service facilities, the scarcity of alternative paths, and the weight of critical corridors; the observation timeliness can be obtained by normalizing the time difference between the current decision time and the most recent effective road state confirmation time. Weights can be determined offline based on historical playback, cross-validation, or emergency experience.

[0086] The posterior entropy can be calculated using the following formula: ; Candidate inspection sections The posterior entropy; S A set of service states; Candidate inspection sections In service s The posterior probability.

[0087] For each candidate road segment in the shortlist of candidate inspection segments, the potential short-term recovery benefits under different potential observation results are evaluated. Let Y be the set of possible observation results obtained after inspecting a candidate road segment. The observation results include at least passage reports, downgraded passage reports, and closure reports. Potential observation results are not the actual state itself, but rather the reported state generated by manual inspection, drone inspection, vehicle-mounted inspection, or fixed monitoring equipment. When the inspection image is unclear, communication is interrupted, or the on-site conditions do not meet the interpretation requirements, Y may also include reports that cannot be determined. The set of potential observation results is predetermined based on the service state set, inspection method, and inspection observation model. The inspection observation model provides the conditional probabilities between different observation reports and potential actual states, based on which the local posterior probability of the candidate road segment under the potential observation result y and the corresponding local posterior offset can be calculated.

[0088] The probability of a potential observation occurring can be calculated using the following formula: ; To inspect candidate road sections The probability of obtaining a potential observation result y after performing an inspection; Let y be the conditional probability of producing the observed result y when the true state is s.

[0089] Let c be the coarse node to which the candidate inspection segment belongs. An activated coarse decision subgraph containing coarse node c and its preset support depth neighborhood is extracted from the offline-constructed fixed coarse decision graph. A bounded support kernel is then constructed using the normalized adjacency matrix of this activated coarse decision subgraph. The activated coarse decision subgraph is obtained by locally cutting around the coarse node to which the candidate inspection segment belongs, based on the fixed coarse decision graph.

[0090] The adjacency matrix of the activated coarse decision subgraph and its normalized form can be expressed as follows: ; .

[0091] To activate coarse nodes in the coarse decision subgraph With coarse nodes The corresponding adjacency matrix elements; For thick edges; B It is an adjacency matrix; I It is the identity matrix; D This is the degree matrix after adding self-loops.

[0092] Based on the normalized propagation weights derived from the bounded support kernel, posterior transmission updates are performed within the support depth neighborhood, ensuring that local observations at the source coarse nodes are propagated in a bounded manner to the relevant recovery neighborhood. After obtaining the bounded support kernel, the kernel vectors corresponding to the source coarse nodes of the candidate inspection segments are extracted, and the kernel values ​​within the support depth neighborhood are normalized to obtain propagation weights. Subsequently, the local posterior offsets of the source coarse nodes are allocated to the neighboring coarse nodes according to the propagation weights, superimposed on the observation and posterior logarithmic values ​​of the neighboring coarse nodes, and normalized to obtain the updated three-state posterior probabilities. The updated coarse node posterior states can be directly used for coarse spatial decision-making, or they can be back-projected to the segment layer according to the coarse node-member segment mapping relationship for subsequent repair priority updates or status display.

[0093] The normalized propagation weight can be calculated using the following formula: ; The normalized propagation weights from the source coarse node c to the neighboring coarse node j; Let c be the element in the c-th row and j-th column of the bounded supporting kernel; Let c be the set of neighborhood coarse nodes of the source coarse node c within the support depth d; u This is the index of the coarse node in the neighborhood set.

[0094] In this way, a single local inspection observation can drive the posterior linkage update of multiple coarse nodes in the relevant neighborhood, thereby more accurately assessing the impact of the inspection action on subsequent recovery decisions without recalculating the entire road network.

[0095] To avoid performing costly simulations for each candidate inspection action at the entire road network scale, this embodiment employs a short-time recovery value reduction estimator to quickly estimate the information value of candidate inspection actions. In the offline phase, given a rolling window length and a fixed fast repair strategy, short-time rolling simulation samples are generated for multiple posterior scenarios. Each simulation sample includes at least one set of low-dimensional summary states of the coarse spatial states. And the corresponding short-term recovery value tag The short-term recovery value can be weighted by indicators such as the recovery volume of key accessibility within the window, the amount of key service demand met, and the utilization efficiency of remaining maintenance resources.

[0096] .

[0097] By using low-dimensional summary state vectors, the key information that determines short-term recovery gains can be preserved in a smaller dimension.

[0098] The short-term recovery information value of each candidate road segment in the shortlist of candidate inspection segments is compared, and the target inspection segment with the highest information value is selected for inspection. After obtaining the inspection results, the corresponding candidate observation branches are replaced with the actual observation results, the posterior state is updated, and the rolling repair priority is calculated for objects that are in a closed or downgraded state and have repair feasibility. The interruption severity item highlights the objects that are currently causing significant service loss and require repair; the repair duration item is used to suppress actions that occupy maintenance resources for too long. Objects to be repaired are selected in descending order of repair priority, and a rolling repair decision is executed under the condition that remaining maintenance resources, construction accessibility, and safety constraints are met.

[0099] This application has at least the following beneficial effects: (1) Define the value of candidate inspection actions as the expected marginal increase in short-term recovery benefits, so that the selection of inspection targets is directly centered around the recovery goal; (2) By constructing a fixed coarse decision graph offline, the road segments related to recovery are compressed into a smaller decision space according to the service coupling relationship and the disaster coupling relationship. In the online stage, only the local coarse decision subgraph is activated, which takes into account both the ability to express the recovery relevance and the efficiency of online computing. (3) By using bounded-supported graph-coupled posterior transmission, local observation results are transformed into posterior linkage updates for restoring relevant neighborhoods, thereby more accurately assessing the impact of inspection actions on subsequent restoration decisions. (4) By replacing the recalculation of each candidate inspection action across the entire road network with a short-term recovery value downgrade evaluator, and unifying inspection decisions and rolling repair decisions under the same recovery benefit target, key accessibility and service capabilities can be restored more quickly under limited inspection budget and maintenance resources.

[0100] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4As shown, the computer device includes a processor, memory, input / output interfaces, and a communication interface. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores the repair priorities and rolling repair decisions for each object to be repaired. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a joint decision-making method for the inspection and restoration of a highway network affected by rainstorms.

[0101] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps of the various method embodiments.

[0102] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the various method embodiments.

[0103] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. The content of this specification should not be construed as a limitation of this application.

Claims

1. A joint decision-making method for highway network inspection and restoration in response to rainstorm disasters, characterized in that, include: Obtain basic data on the highway network to be restored; Construct a fixed coarse decision map of the road sections to be restored in the road network to be restored; Based on the basic data, determine the posterior probability that each restored road segment belongs to a different service state; Based on the posterior probability, the restored road sections are pre-screened to obtain a short list of candidate inspections; Based on the observation model and the observation results of different potential inspections, the local posterior offset of each candidate inspection segment in the shortlist of candidate inspections is determined. Extract the activated coarse decision subgraph containing candidate inspection segments from the fixed coarse decision graph, construct a bounded support kernel using the adjacency matrix of the activated coarse decision subgraph, and propagate the local posterior offset to the neighborhood within the activated coarse decision subgraph based on the bounded support kernel to obtain the updated coarse spatial posterior state. After converting the updated coarse space posterior state into a summary state vector, it is input into the short-time recovery value down-order evaluator to obtain the updated short-time recovery value. The expected value of the difference between the updated short-term recovery value and the baseline short-term recovery value before observation is obtained to obtain the short-term recovery information value of each candidate inspection segment. The candidate road sections with the highest short-term recovery information value are selected from the shortlist of candidate inspections and inspected. The repair priority of each object to be repaired is updated based on the inspection results, and rolling repair decisions are made according to the updated repair priority under the constraints of maintenance resources.

2. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, The basic data includes: highway network data, historical disaster point data, hydrological and meteorological forcing data, road condition confirmation data, recovery resource data, and key service demand data.

3. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, Construct a fixed coarse decision map of the road segments to be restored in the road network to be restored, including: Construct a directed road segment map based on the road network data in the basic data; A subset of restored road segments is selected from the directed road segment map. The restored road segment subset meets the following conditions: it is located on the critical origin-end route or emergency access, intersects with the historical high-risk disaster corridor, serves as one of the alternative routes to bypass the historical high-risk disaster corridor, or has an impact on the satisfaction of critical service needs that is greater than the impact threshold; the risk disaster incidence rate of the high-risk disaster corridor is greater than the risk threshold. Determine the service correlation and disaster correlation of road segment pairs in the restored road segment subset, and merge the service correlation and disaster correlation to obtain static pairwise similarity; While maintaining the connectivity constraints of the directed road segment graph, agglomerative clustering is performed on the recovered road segment subset based on static pairwise similarity to obtain multiple coarse nodes; Coarse edges are constructed based on the original adjacency relationship of the restored road segments within the coarse nodes or when the average similarity between coarse nodes is greater than a preset similarity threshold, thus forming a fixed coarse decision graph.

4. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 3, characterized in that, The process of determining service relevance includes: obtaining the set of origin-destination paths, the set of facilities served, and the set of corridors affected by the two road segments in the road segment pair; calculating the overlap of the origin-destination path sets, the facility sets, and the corridor sets for the two road segments respectively; and performing a weighted summation of the overlap of the origin-destination path sets, the facility sets, and the corridor sets to obtain the service relevance. The process of determining the disaster correlation degree includes: calculating the consistency of dominant disaster type, spatial distance attenuation term, drainage unit consistency, and route category consistency between the two road segments; and then normalizing and weighting the consistency of dominant disaster type, spatial distance attenuation term, drainage unit consistency, and route category consistency to obtain the disaster correlation degree.

5. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, Based on the basic data, the posterior probability of each restored road segment belonging to different service states is determined, including: Based on the highway network data, historical disaster point data, and key service demand data in the basic data, a priori disaster conditions are constructed under different service states; the priori disaster conditions include the prior risk values ​​of each restored road section; Based on the highway network data, hydrological and meteorological forcing data, road condition confirmation data, and historical disaster point data in the basic data, the posterior logarithmic values ​​of each restored road segment under different service states are calculated using a posterior inference model; different service states include traffic status, downgraded traffic status, and closed status. The posterior logarithmic values ​​of different service states are superimposed with the prior field of disaster conditions under their respective service states and then normalized to obtain the posterior probability of each restored road segment belonging to different service states.

6. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, Based on posterior probabilities, the restored road sections are pre-screened to obtain a shortlist of candidate inspection sections, including: Use the posterior probability of the closed state as the closed state probability; Determine the posterior entropy based on the posterior probability; Determine the time interval or normalized result of the time interval between the current time and the most recent valid road status confirmation time of the restored road section, and use it as the observation time limit; The closed-state probability term, posterior entropy, service criticality, and observation timeliness are weighted and summed to obtain the pre-screening score for each restored road segment; According to the preset screening rules, road sections with pre-screened scores not less than the score threshold are selected for restoration, or, according to the pre-screened scores from high to low, no more than a preset number of road sections are selected to form a shortlist of candidate inspection sections.

7. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, Based on the observation model and observation results of different potential inspections, the local posterior offset of each candidate inspection segment in the shortlist of candidate inspections is determined, including: Based on the conditional probability between the reported state and the potential true state corresponding to the inspection results in different potential inspection observation results, the inspection observation model is adopted to determine the local posterior logarithmic value after observation. The difference between the local posterior logarithmic value after observation and the local posterior logarithmic value before observation is calculated to obtain the local posterior offset of each candidate inspection segment in the shortlist of candidate inspection segments.

8. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, Based on the bounded support kernel, the local posterior offset is propagated to the neighborhood within the activated coarse decision subgraph to obtain the updated coarse spatial posterior state, including: Project the local posterior offset onto the corresponding coarse node in the activated coarse decision subgraph; Extract the kernel vector corresponding to the coarse node in the bounded support kernel, and normalize the kernel vector in the neighborhood of the preset support depth to obtain the normalized weight. Multiply the local posterior offset by the normalized weight of the corresponding neighboring coarse nodes to obtain the posterior log increment of each neighboring coarse node. The posterior logarithmic increment is superimposed onto the local posterior logarithmic values ​​of each neighborhood coarse node before observation, and then normalized to obtain the updated coarse spatial posterior state of each coarse node.

9. The joint decision-making method for highway network inspection and restoration caused by rainstorms as described in claim 1, characterized in that, The summary state vector includes one or more of the following: the closure quality values ​​of the top k coarse nodes with the highest probability of activating the coarse decision subgraph; the degradation quality values ​​of the top k coarse nodes with the highest probability of activating the degraded accessibility state; the current key accessibility indicators; the current proportion of served demand; the remaining maintenance resource capacity; the average posterior entropy of activating the coarse decision subgraph; the dispersion of the posterior entropy of activating the coarse decision subgraph; and the average closure quality of activating the coarse decision subgraph.

10. The joint decision-making method for highway network inspection and restoration caused by rainstorms according to claim 1, characterized in that, The repair priority is calculated by weighting the predicted improvement in critical accessibility of the repair action, the current severity of the interruption of the object to be repaired, and the expected repair time. Alternatively, the repair priority can be calculated by weighting the predicted improvement in critical accessibility caused by the repair action, the degree of service loss of the object to be repaired, and the expected repair time.

11. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the joint decision-making method for rainstorm-induced road network inspection and restoration as claimed in any one of claims 1-10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the joint decision-making method for the inspection and restoration of highway networks affected by rainstorms, as described in any one of claims 1-10.

13. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the joint decision-making method for the inspection and restoration of highway networks affected by rainstorms, as described in any one of claims 1-10.

Images