Bridge structure water damage safety state monitoring method, device and system based on bayesian network

Abstract

The application provides a bridge structure water damage safety state monitoring method, device and system based on a Bayesian network, acquires environmental action data and structural response data of a region where a bridge is located, combines historical water damage event records and bridge structure damage evolution rules, constructs an initial Bayesian network topology for describing the relationship between environmental factors and structural damage, inputs the environmental action data and the structural response data into the initial Bayesian network topology for parameter learning, adjusts the conditional probability distribution of each node in the network, generates an updated network topology with real-time adjustment capability, and based on this, fuses the real-time collected environmental action data and the structural response data, performs network state inference processing, obtains a state inference result reflecting the current safety level of the bridge structure, and according to the state inference result, generates a safety state evaluation result containing structural damage risk sorting and key risk factor identification.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, device, and system for monitoring the safety status of bridge structures damaged by water based on Bayesian networks. Background Technology

[0002] Bridge structural water damage safety status monitoring involves continuously monitoring and analyzing the mechanical response and deformation characteristics of bridge structures under environmental influences such as water erosion and precipitation infiltration to assess structural damage risks and provide early warnings of potential safety hazards. Currently, most monitoring methods rely on traditional mechanical analysis or single data-driven models. Typically, after collecting structural stress-strain or displacement response data, damage status is determined by setting preset thresholds. When probabilistic models are introduced, their network topologies are often statically constructed based on historical statistical data, with fixed inter-node relationships and parameter updates dependent on periodic manual intervention. This fails to fully integrate the physical mechanisms of bridge structural damage evolution with the dynamic characteristics of actual environmental effects. These approaches suffer from unclear causal relationships between environmental factors and structural damage, difficulty in adapting model parameters to changes in environmental conditions, and insufficient integration of real-time monitoring data with the network inference process, affecting the accuracy of assessing the current safety level of bridges and the effectiveness of identifying risk factors. Summary of the Invention

[0003] In view of this, the present invention provides a method, device and system for monitoring the safety status of bridge structures damaged by water based on Bayesian networks.

[0004] The technical solution of this invention is implemented as follows:

[0005] On one hand, embodiments of the present invention provide a method for monitoring the safety status of bridge structures damaged by water based on Bayesian networks. The method includes: acquiring environmental action data and structural response data of the area where the bridge is located; the environmental action data includes water flow monitoring information reflecting hydrodynamic conditions and precipitation monitoring information reflecting meteorological change characteristics; the structural response data includes stress-strain monitoring information characterizing the mechanical behavior of bridge components and displacement monitoring information characterizing overall deformation characteristics; combining historical water damage event records and bridge structural damage evolution patterns, constructing an initial Bayesian network topology to describe the relationship between environmental factors and structural damage; the initial Bayesian network topology includes a set of directed edges reflecting the transmission path of environmental actions and a node hierarchy representing the structural damage state; inputting the environmental action data and structural response data into the initial Bayesian network topology for parameter learning, adjusting the conditional probability distribution of each node in the network, and generating an updated network topology with real-time adjustment capabilities; based on the updated network topology, fusing the real-time acquired environmental action data and structural response data, performing network state reasoning processing to obtain a state reasoning result reflecting the current safety level of the bridge structure; and generating a safety status assessment result including structural damage risk ranking and identification of key risk factors based on the state reasoning result.

[0006] On the other hand, embodiments of the present invention provide a bridge structure water damage safety status monitoring device. The device includes: a data acquisition module for acquiring environmental impact data and structural response data of the area where the bridge is located. The environmental impact data includes water flow monitoring information reflecting hydrodynamic conditions and precipitation monitoring information reflecting meteorological change characteristics. The structural response data includes stress-strain monitoring information characterizing the mechanical behavior of bridge components and displacement monitoring information characterizing overall deformation characteristics. A network construction module is used to combine historical water damage event records and bridge structural damage evolution patterns to construct an initial Bayesian network topology describing the relationship between environmental factors and structural damage. The initial Bayesian network topology includes data reflecting environmental factors and structural damage. The system comprises: a set of directed edges representing the action transmission path and a node hierarchy representing the structural damage state; a network update module, used to input the environmental action data and structural response data into the initial Bayesian network topology for parameter learning, adjust the conditional probability distribution of each node in the network, and generate an updated network topology with real-time adjustment capabilities; a state reasoning module, used to perform network state reasoning processing based on the updated network topology, integrating real-time collected environmental action data and structural response data, to obtain a state reasoning result reflecting the current safety level of the bridge structure; and a state assessment module, used to generate a safety state assessment result including structural damage risk ranking and key risk factor identification based on the state reasoning result.

[0007] Thirdly, the present invention provides a computer system including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the program to implement the steps in the above-described method.

[0008] This invention helps improve the accuracy and reliability of monitoring the safety status of bridge structures under water damage. By collaboratively collecting environmental action data and structural response data, it provides a more comprehensive data foundation for analyzing the state of bridge structures under water damage, avoiding the limitations of single-type data in reflecting the complexity of water damage. An initial Bayesian network topology is constructed by combining historical water damage event records and bridge structural damage evolution patterns, enabling the network topology to better reflect the intrinsic relationship between environmental factors and structural damage, thus improving the model's interpretability. By adjusting the conditional probability distribution of nodes through parameter learning and generating an updated network topology, the network can adapt to changes in bridge structure and environmental conditions, maintaining good predictive capabilities. Network state inference based on the updated network topology and real-time collected data helps to obtain timely and accurate state inference results reflecting the current safety level of the bridge structure. Based on the state inference results, a safety status assessment result is generated, including structural damage risk ranking and identification of key risk factors. This not only clarifies the risk status of the bridge structure but also helps to trace the main causes of risk, providing more valuable reference information for bridge safety management and maintenance decisions. Attached Figure Description

[0009] Figure 1 This is a schematic diagram illustrating the implementation process of a bridge structure water damage safety status monitoring method based on Bayesian networks, provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of the composition of a bridge structure water damage safety status monitoring device provided in an embodiment of the present invention.

[0011] Figure 3 This is a schematic diagram of the hardware entity of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0012] This invention provides a method for monitoring the safety status of bridge structures damaged by water based on Bayesian networks. This method can be executed by a processor of a computer system. The computer system can refer to devices with corresponding data processing capabilities, such as servers, laptops, tablets, and desktop computers.

[0013] Figure 1 This is a schematic diagram illustrating the implementation process of a bridge structure water damage safety status monitoring method based on Bayesian networks, as provided in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0014] Step S100: Obtain environmental action data and structural response data for the area where the bridge is located. The environmental action data includes water flow monitoring information reflecting hydrodynamic conditions and precipitation monitoring information reflecting meteorological change characteristics. The structural response data includes stress and strain monitoring information characterizing the mechanical behavior of bridge components and displacement monitoring information characterizing the overall deformation characteristics.

[0015] Environmental impact data refers to data on external environmental factors affecting bridge structures. Water flow monitoring information reflects the hydrodynamic conditions around the bridge, such as flow rate, volume, and direction. To obtain water flow monitoring information, current meters and water level gauges can be installed at appropriate locations in the river near the bridge. Current meters measure water flow velocity in real time; by collecting and analyzing velocity data at different locations and times, the distribution and variation patterns of water flow velocity can be understood. Water level gauges measure water level; combined with information such as river topography and water flow velocity, the flow rate can be calculated.

[0016] Rainfall monitoring information reflects the meteorological characteristics of the area where the bridge is located, covering rainfall amount, intensity, and duration. Heavy rainfall can cause river levels to rise and soil moisture content to increase, thus affecting the structural safety of the bridge. For example, torrential rain can cause floods, increasing the impact and scouring force of water flow on the bridge; prolonged rainfall can saturate the soil, reducing the bearing capacity of the bridge foundation. To obtain rainfall monitoring information, rain gauges can be installed around the bridge to measure rainfall. Statistical analysis of rainfall data can yield information such as rainfall intensity and duration. Structural response data is the feedback information generated by the bridge structure under environmental influences. Stress-strain monitoring information displays the mechanical performance of bridge components, reflecting the stress and strain conditions of the components under load. Stress is the ratio of internal force to cross-sectional area of ​​a component, and strain is the ratio of deformation under load to its original size. By analyzing stress-strain monitoring information, the stress state and safety performance of bridge components can be understood. To obtain stress-strain monitoring information, strain gauges and fiber optic sensors can be installed at key locations on bridge components. Strain gauges can measure the surface strain of components, and by collecting and processing the strain data, the stress of the components can be calculated. Fiber optic sensors have high accuracy and strong anti-interference capabilities, enabling more precise measurement of the stress and strain of components.

[0017] Displacement monitoring information reflects the overall deformation characteristics of a bridge, including vertical and horizontal displacement. This information is crucial for assessing the overall stability and safety of a bridge. For example, during long-term use, bridges may experience displacement due to factors such as foundation settlement and temperature changes; excessive displacement can affect the normal use and safety of the bridge. To obtain displacement monitoring information, measuring equipment such as total stations and GPS can be used. Total stations measure the angles and distances at different points on the bridge and calculate coordinate changes to determine displacement. GPS can monitor bridge displacement in real time, offering advantages such as high precision and all-weather operation.

[0018] Step S200: Combining historical records of water damage events and the evolution of bridge structural damage, construct an initial Bayesian network topology to describe the relationship between environmental factors and structural damage. The initial Bayesian network topology includes a set of directed edges that reflect the transmission path of environmental effects and a node hierarchy that represents the state of structural damage.

[0019] Historical records of flood damage events contain detailed information about past bridge flooding incidents, such as the time and location of the event, the environmental conditions at the time, and the extent of damage to the bridge structure. The law of bridge structural damage evolution describes the process and characteristics of damage development in bridge structures under environmental influences, involving knowledge from multiple fields such as structural mechanics and materials science.

[0020] The initial Bayesian network topology is a probabilistic graphical model used to describe the relationship between environmental factors and structural damage. It consists of nodes and directed edges. Nodes are divided into environmental factor nodes and structural damage nodes, and directed edges represent the causal relationships between environmental factors and structural damage. The set of directed edges reflects the transmission path of environmental effects, that is, the path through which environmental factors influence the damage state of the bridge structure through a series of physical processes. The hierarchical relationship of nodes reflects different levels and stages of structural damage, such as different grades from minor damage to severe damage.

[0021] In one implementation, step S200 may specifically include the following steps S210 to S260:

[0022] Step S210: Perform spatiotemporal correlation mining on the correlation characteristics between environmental factors and structural damage in historical flood damage event records to obtain a set of causal correlation descriptions of environmental factors and structural damage at different time stages and spatial locations.

[0023] Spatiotemporal correlation mining is an operation that analyzes and mines the correlation characteristics between environmental factors and structural damage in historical flood damage event records across time and space. Historical flood damage event records contain a wealth of information on environmental factors and structural damage, which exhibit specific distributions in time and space. For example, rainfall and water flow velocity vary in different seasons, and the structural damage conditions of different parts of a bridge also differ. Through spatiotemporal correlation mining, this information can be integrated and analyzed to determine the spatiotemporal correlation patterns between environmental factors and structural damage.

[0024] When performing spatiotemporal correlation mining, data mining algorithms, such as association rule mining algorithms and cluster analysis algorithms, can be used. Association rule mining algorithms can identify frequent correlation patterns between environmental factors and structural damage. For example, they can find that when precipitation exceeds a certain threshold and water flow velocity is within a preset range, the probability of bridge foundation scour increases significantly. Cluster analysis algorithms can cluster historical water damage event records according to temporal and spatial characteristics to identify similar event patterns, thereby better understanding the causal relationship between environmental factors and structural damage.

[0025] Through spatiotemporal correlation mining, a set of causal relationship descriptions between environmental factors and structural damage at different time stages and spatial locations can be obtained. This set contains a series of causal relationship descriptions, each explaining how environmental factors lead to structural damage at a specific time stage and spatial location. For example, during a certain period of a certain season, increased water flow velocity downstream of the bridge can lead to increased horizontal displacement of the bridge piers.

[0026] Step S220: Screen the correlation strength between environmental factors and structural damage in the causal correlation description set to generate a correlation strength sequence between environmental factors and structural damage. The values ​​in the correlation strength sequence are related to the frequency and duration of the co-occurrence of environmental factors and structural damage.

[0027] The strength of the correlation reflects the degree of association between environmental factors and structural damage, indicating the magnitude of the impact of environmental factors on structural damage. Within a causal correlation description set, the strength of the correlation between different environmental factors and structural damage varies. For example, the correlation strength between water flow velocity and pier displacement may be greater than the correlation strength between precipitation and pier displacement.

[0028] Filtering is the process of selecting meaningful association strength information from a set of causal association descriptions. Since the set may contain a large amount of association information, some associations may be accidental or unimportant, so filtering is needed to obtain more accurate and useful association strength information.

[0029] A correlation strength sequence is a numerical sequence of correlation strengths arranged in a certain order, where each value represents the correlation strength between an environmental factor and structural damage. The value is related to the frequency and duration of the co-occurrence of the environmental factor and structural damage. If an environmental factor and structural damage frequently occur simultaneously and for a long period, the correlation strength between them is greater.

[0030] Statistical analysis methods, such as frequency statistics and correlation analysis, can be used during the screening process. Frequency statistics calculate the frequency of co-occurrence of environmental factors and structural damage, while correlation analysis calculates the correlation coefficient between them to assess the strength of the association. Based on the analysis results, the association information between environmental factors and structural damage with strong correlations is screened out, generating a correlation strength sequence.

[0031] In one implementation, step S220 may specifically include the following steps S221 to S226:

[0032] Step S221: Perform multi-feature decomposition on the environmental action parameters in the causal association description set, and extract the impact and scour features from the water flow monitoring information, and the intensity and duration features from the precipitation monitoring information.

[0033] Environmental impact parameters are specific parameters of environmental factors within a causal relationship description set, such as the parameters in water flow and precipitation monitoring information. Multi-feature decomposition (MFD) is the process of decomposing environmental impact parameters into multiple features with different physical meanings. The impact feature in water flow monitoring information reflects the magnitude of the impact force of water flow on the bridge structure, and is related to factors such as water flow velocity and flow rate. For example, the faster the water flow, the greater the impact force on the bridge piers. The scouring feature reflects the scouring effect of water flow on the bridge foundation, and is related to water flow velocity, flow direction, and sediment content. For example, a high sediment content in the water flow results in a stronger scouring effect on the bridge foundation. The intensity feature in precipitation monitoring information reflects the magnitude of precipitation intensity, usually expressed as precipitation per unit time. The duration feature reflects the duration of precipitation and has a significant impact on soil moisture content and river water levels. For example, prolonged precipitation may lead to rising river water levels, increasing the water pressure on the bridge.

[0034] When performing multi-feature decomposition processing, for water flow monitoring information, impact and scour characteristics can be extracted by analyzing data such as water flow velocity and flow rate. For example, calculating the rate of change of water flow velocity reflects impact characteristics, and analyzing changes in sediment content in the water flow reflects scour characteristics. For precipitation monitoring information, precipitation amount and precipitation time data are processed to extract intensity and duration characteristics. For example, calculating the precipitation amount per unit time yields intensity characteristics, and statistically analyzing the duration of precipitation yields duration characteristics.

[0035] Step S222: Perform multi-dimensional segmentation processing on the structural damage indicators in the causal correlation description set, decompose the stress-strain monitoring information into instantaneous response indicators and cumulative effect indicators, and decompose the displacement monitoring information into local deformation indicators and overall deformation indicators.

[0036] Structural damage indices are various indicators that describe the damage state of bridge structures, such as stress-strain and displacement monitoring information. Multi-dimensional segmentation involves dividing structural damage indices according to different dimensions to comprehensively describe the damage situation of the bridge structure.

[0037] Instantaneous response indices in stress-strain monitoring information reflect the bridge structure's response to instantaneous stress and are related to the stress and strain at the moment of loading. For example, the instantaneous stress and strain values ​​generated in bridge components when a vehicle crosses a bridge are instantaneous response indices. Cumulative effect indices reflect the cumulative damage to the bridge structure under long-term stress and are related to the cumulative effect of stress and strain. For example, during long-term use of a bridge, repeated loading and unloading cause cumulative stress and strain within the components, leading to fatigue damage; cumulative effect indices can reflect this damage.

[0038] Local deformation indices in displacement monitoring data reflect the deformation of local parts of the bridge structure, such as local tilting of piers or local cracks in the beams. Overall deformation indices reflect the overall displacement of the bridge structure, such as vertical settlement and horizontal displacement.

[0039] When performing multi-dimensional segmentation, stress-strain and displacement monitoring information is analyzed and processed based on structural mechanics principles and data analysis methods. For stress-strain monitoring information, data from different time periods are analyzed to separate instantaneous response indicators and cumulative effect indicators. For example, analyzing stress-strain data when a vehicle passes over a bridge yields instantaneous response indicators, while cumulative calculations are performed on long-term stress-strain data to obtain cumulative effect indicators. For displacement monitoring information, displacement data from different parts of the bridge are analyzed to differentiate between local deformation indicators and overall deformation indicators. For example, analyzing displacement data from piers and beams separately yields corresponding indicators.

[0040] Step S223: Calculate the correlation strength values ​​between each environmental action parameter and the structural damage index based on a multi-scale sliding window. The length of the sliding window is adjusted according to the duration characteristics of the environmental action.

[0041] The duration characteristics of environmental actions refer to the duration and variation patterns of environmental actions (such as water flow and precipitation). Different environmental actions have different duration characteristics; for example, heavy rainstorms have short durations, while floods have long durations. The sliding window length needs to be adjusted according to the duration characteristics of environmental actions to accurately capture the correlation between environmental actions and structural damage.

[0042] When calculating the correlation strength value, the environmental action parameters and structural damage index data are first arranged in chronological order. Then, a multi-scale sliding window is slid across the data sequence, and correlation analysis is performed on the data within the window to calculate the correlation strength value. For example, methods such as the Pearson correlation coefficient can be used. For instance, for water flow velocity and pier displacement parameters, it is assumed that the water flow velocity and pier displacement data are recorded over time. A sliding window is set, with the window length adjusted according to the duration of the water flow action. The window is slid across the data sequence, and the Pearson correlation coefficient of the water flow velocity and pier displacement data within the window is calculated each time, obtaining the correlation strength value for that window. By continuously sliding the window, a series of correlation strength values ​​can be obtained.

[0043] Step S224: Perform time series alignment processing on the calculated correlation strength values ​​so that the correlation strength values ​​of different environmental action parameters and structural damage indicators are on the same time axis.

[0044] Time series alignment is the process of aligning the correlation strength values ​​of different environmental parameters and structural damage indicators in chronological order. Since the sampling times and data lengths of different environmental parameters and structural damage indicators may vary, the calculated correlation strength values ​​may not lie on the same time axis. Interpolation and resampling methods can be used for time series alignment. Interpolation methods estimate the correlation strength values ​​at missing time points based on known correlation strength values. For example, linear interpolation estimates the value at the intermediate time point based on the correlation strength values ​​of two adjacent time points. Resampling methods unify correlation strength values ​​with different sampling frequencies to the same sampling frequency. For example, high-frequency correlation strength values ​​are downsampled to have the same time interval as low-frequency values.

[0045] Step S225: Generate a correlation strength sequence based on the correlation strength values ​​after time series alignment. Each element in the correlation strength sequence corresponds to the degree of correlation between environmental factors and structural damage within the time window.

[0046] A correlation strength sequence is a numerical sequence of correlation strengths arranged chronologically, reflecting the degree of correlation between environmental factors and structural damage within different time windows. After aligning the time series, the correlation strength values ​​are on the same time axis, which can be used to generate a correlation strength sequence.

[0047] When generating the correlation strength sequence, the correlation strength values ​​after time series alignment are arranged in chronological order, with each value corresponding to a time window. For example, the correlation strength values ​​between a certain environmental parameter and a structural damage index, arranged in chronological order, are [correlation strength value 1, correlation strength value 2, correlation strength value 3, ...], which constitute the correlation strength sequence. The first value represents the degree of correlation within the first time window, the second value represents the degree of correlation within the second time window, and so on. The correlation strength sequence can be used for subsequent analysis and processing, such as statistical analysis to determine the maximum, minimum, and average values ​​of the correlation strength, assess the overall correlation between environmental factors and structural damage, and perform trend analysis to observe the changing trend of the correlation strength over time and predict future correlation conditions.

[0048] Step S226: Perform hierarchical clustering on the association strength sequences, merge sequence segments with similar association patterns, and generate a set of association strength sequences representing different association patterns.

[0049] Hierarchical clustering is a data clustering method that forms cluster structures at different levels by continuously merging similar data objects. In this step, hierarchical clustering is performed on the association strength sequences, merging sequence fragments with similar association patterns to generate a set of association strength sequences representing different association patterns.

[0050] Association patterns refer to the characteristics and regularities of association strength sequences over time. Different association patterns may reflect different mechanisms of action between environmental factors and structural damage. For example, some association patterns show that the association strength gradually increases over time, while others fluctuate significantly within a certain period.

[0051] In hierarchical clustering, a similarity metric for sequence segments is first defined, such as Euclidean distance or Manhattan distance. Then, based on this similarity metric, the similarity between sequence segments in the association strength sequence is calculated. Next, sequence segments with high similarity are merged into a single cluster, and this process is repeated until different levels of clustering structures are formed. Finally, through hierarchical clustering, a set of association strength sequences representing different association patterns is obtained. The association patterns of the association strength sequences in each set are similar, and this can be used to further analyze the relationship between environmental factors and structural damage under different association patterns.

[0052] Step S230: Divide the node types of the initial Bayesian network topology according to the correlation strength sequence to obtain the set of environmental factor nodes and the set of structural damage nodes. The set of environmental factor nodes corresponds to various influencing factors in the environmental action data, and the set of structural damage nodes corresponds to various damage states in the structural response data.

[0053] The correlation strength sequence reflects the degree of correlation between environmental factors and structural damage. Based on this sequence, the node types of the initial Bayesian network topology can be classified, mainly into environmental factor nodes and structural damage nodes. The set of environmental factor nodes encompasses various influencing factors in environmental action data, such as water flow velocity and flow rate in water flow monitoring information, and precipitation amount and intensity in precipitation monitoring information. These nodes represent external environmental factors that may affect the bridge structure. The set of structural damage nodes contains various damage states in structural response data, such as instantaneous response indicators and cumulative effect indicators in stress-strain monitoring information, and local deformation indicators and overall deformation indicators in displacement monitoring information. These nodes represent the potential damage to the bridge structure under environmental influences. When classifying node types, the strong correlation between environmental factors and structural damage is determined based on the magnitude and trend of the correlation strength sequence. For example, if the correlation strength sequence shows a strong correlation between water flow velocity and pier displacement, water flow velocity can be designated as an environmental factor node, and pier displacement as a structural damage node. By rationally classifying node types, the foundation for constructing the initial Bayesian network topology is laid.

[0054] Step S240: By analyzing the physical process of environmental factors acting on structural damage, determine the direction of the directed edges between the set of environmental factor nodes and the set of structural damage nodes, and generate a set of directed edges containing direct action paths and indirect transmission paths.

[0055] The physical process by which environmental factors affect structural damage refers to the occurrence and development of bridge structural damage caused by environmental factors through a series of physical actions. For example, increased water flow velocity increases the impact force of the water flow on the bridge piers, leading to pier displacement and foundation scour—this is a physical process. The directed edge direction indicates the direction of the causal relationship between environmental factors and structural damage, i.e., how environmental factors affect structural damage. A direct action path is the path through which environmental factors directly cause structural damage, such as the impact force of water flow directly acting on the bridge piers, causing displacement. An indirect transmission path is the path through which environmental factors indirectly cause structural damage through intermediate physical effects, such as increased water flow velocity causing river water levels to rise, increasing water pressure on the bridge foundation, and water pressure leading to foundation settlement.

[0056] In one implementation, step S240 may specifically include the following steps S241 to S246:

[0057] Step S241: Perform an action time analysis on the environmental action parameters corresponding to each node in the set of environmental factor nodes to determine the start time and duration of each environmental factor acting on the bridge structure.

[0058] The environmental action parameters corresponding to each node in the environmental factor node set are specific physical quantities related to the environmental factors, such as water flow velocity and precipitation. Action time analysis is the process of determining the duration for which these parameters act on the bridge structure.

[0059] The start time is the point at which environmental factors begin to affect the bridge structure. For example, the point at which water flow velocity begins to increase and exert impact on the bridge piers. The duration is the length of time that environmental factors affect the bridge structure. For example, the period during which water flow velocity continues to increase and exert impact on the bridge piers.

[0060] When conducting action time analysis, the analysis is based on monitoring data of environmental action parameters and response data of the bridge structure. For example, analyzing water flow velocity monitoring data, the time point when the velocity begins to increase is determined as the start time; the period of continuous velocity increase is analyzed as the duration. The action time can also be verified by combining stress-strain and displacement monitoring information of the bridge structure. For example, after the water flow velocity increases, the stress-strain and displacement data of the bridge piers begin to change, and this time point can be confirmed as the start time.

[0061] Step S242: Perform occurrence time analysis on the structural damage index corresponding to each node in the set of structural damage nodes to determine the initial time and development time of each structural damage.

[0062] The structural damage indicators corresponding to each node in the structural damage node set are indices describing the damage state of the bridge structure, such as stress-strain and displacement monitoring information. Occurrence time analysis is the process of determining when the structural damage occurred corresponding to these indicators. Initial time is the point at which structural damage begins to appear. For example, the point at which the pier displacement begins to exceed the normal range. Development time is the time it takes for structural damage to develop from its initial state to a certain extent. For example, the time it takes for the pier displacement to develop from a small displacement to a larger displacement.

[0063] When conducting time-of-occurrence analysis, the analysis is based on monitoring data of structural damage indicators. For example, analyzing bridge pier displacement monitoring data, the initial time is determined when the displacement begins to exceed the normal range; analyzing the trend of displacement over time, the time period from the initial state to the larger displacement is determined as the development time. The analysis of the duration of environmental factors can also be combined to determine whether the occurrence and development of structural damage are related to environmental factors. For example, if bridge pier displacement begins and develops shortly after an increase in water flow velocity, it can be inferred that the two are related.

[0064] Step S243: By comparing the time difference between the time of environmental factor action and the time of structural damage, determine the direction of the causal relationship between environmental factors and structural damage.

[0065] The time difference is the difference between the initial time when environmental factors begin to affect a bridge structure and the initial time when structural damage occurs. Comparing this time difference can determine the direction of the causal relationship between environmental factors and structural damage. If the onset time of the environmental factor's effect is earlier than the initial time of structural damage, and structural damage begins to appear and develop within the duration of the environmental factor's effect, then the environmental factor can be considered the cause of the structural damage, and the causal relationship points from the environmental factor to the structural damage. For example, if the increase in water flow velocity occurs earlier than the occurrence of pier displacement, and the pier displacement increases during the period of continuous increase in water flow velocity, then it can be determined that the increase in water flow velocity caused the pier displacement.

[0066] If the onset of environmental factors is later than the initial time of structural damage, it can be assumed that there may be no direct causal relationship between the two, or that the structural damage was caused by other factors. By comparing the time difference, the direction of the causal relationship between environmental factors and structural damage can be accurately determined, providing a basis for determining the direction of the directed edge.

[0067] Step S244: For environmental factor nodes and structural damage nodes with direct causal relationship, establish directed edges that directly connect them to form a direct action path.

[0068] Direct causality refers to the relationship where environmental factors directly cause structural damage. For example, the impact force of water flow directly acts on a bridge pier, causing displacement. After determining the direct causal relationship between environmental factor nodes and structural damage nodes, a directly connected directed edge is established, pointing from the environmental factor node to the structural damage node, representing the direct impact of the environmental factor on the structural damage. The direct action path is formed by directly connected directed edges, reflecting the process by which environmental factors directly cause structural damage. For example, if there is a direct causal relationship between the water flow velocity node and the bridge pier displacement node, a directed edge is established pointing from the water flow velocity node to the bridge pier displacement node, forming a direct action path. Establishing direct action paths can visually present the direct causal relationship between environmental factors and structural damage, helping to analyze and predict the water damage safety status of bridge structures. Through direct action paths, it can be clearly identified which environmental factors directly affect which structural damages, providing a basis for taking protective measures.

[0069] Step S245: For environmental factor nodes and structural damage nodes with indirect causal relationship, establish directed edge chains with indirect connection by introducing intermediate transmission nodes to form indirect transmission paths. The intermediate transmission nodes characterize the intermediate physical effects of environmental factors on the structural damage process.

[0070] Indirect causation refers to the relationship where environmental factors indirectly cause structural damage through intermediate physical effects. For example, increased water flow velocity causes river water levels to rise, which in turn increases water pressure on bridge foundations, leading to foundation settlement. Intermediate transmission nodes are the points that act as conduits in the process of environmental factors affecting structural damage, representing intermediate physical effects. In the example above, the river water level and the water pressure on the bridge foundations are intermediate transmission nodes, representing the increase in water level caused by increased water flow velocity and the increase in foundation water pressure caused by the increased water level, respectively.

[0071] After determining that there is an indirect causal relationship between environmental factor nodes and structural damage nodes, intermediate transmission nodes are introduced to establish a directed edge chain of indirect connections. The directed edge direction points from the environmental factor node to the intermediate transmission node, and then from the intermediate transmission node to the structural damage node, forming an indirect transmission path. For example, if there is an indirect causal relationship between the water flow velocity node and the bridge foundation settlement node, the river water level node and the bridge foundation water pressure node are introduced as intermediate transmission nodes, and corresponding directed edges are established to form an indirect transmission path.

[0072] Establishing indirect transmission paths can comprehensively reveal the causal relationship between environmental factors and structural damage, taking into account the influence of intermediate physical effects. Through indirect transmission paths, we can gain a deeper understanding of how environmental factors affect structural damage at different points in time via intermediate physical effects, providing support for accurately assessing the water damage safety status of bridge structures.

[0073] Step S246: Perform path redundancy detection on the directed edges in the direct action path and the indirect transmission path, remove redundant directed edges that can be replaced by other paths, and generate a set of directed edges.

[0074] Path redundancy detection is the process of examining directed edges in direct and indirect transmission paths to identify redundant directed edges that can be replaced by other paths. Redundant directed edges are those edges that, when representing the causal relationship between environmental factors and structural damage, exist where other paths can achieve the same effect.

[0075] When constructing the initial Bayesian network topology, redundant directed edges may exist, increasing network complexity and affecting the accuracy of inference results. Therefore, path redundancy detection is necessary to remove redundant directed edges. Graph theory algorithms, such as the shortest path algorithm and reachability analysis algorithm, can be used for path redundancy detection. The shortest path algorithm can determine the shortest path between environmental factor nodes and structural damage nodes. If a directed edge is not on the shortest path but can reach the same structural damage node through other paths, the edge may be redundant. The reachability analysis algorithm can check whether environmental factor nodes can reach structural damage nodes through other paths. If so, the directed edge connecting the two nodes may be redundant. By removing redundant directed edges through path redundancy detection, a concise and effective set of directed edges is generated. This set accurately represents the causal relationship and action path between environmental factors and structural damage, providing a reasonable basis for constructing the initial Bayesian network topology.

[0076] Step S250: Construct a hierarchical transmission network from environmental factor nodes to structural damage nodes based on a set of directed edges. Set conditional dependency parameters between nodes at the same level according to the dependency relationship of damage evolution.

[0077] A hierarchical transmission network is a network with a hierarchical structure, consisting of environmental factor nodes, structural damage nodes, and directed edges. It is used to represent the development of structural damage caused by environmental factors through a transmission process. In a hierarchical transmission network, environmental factor nodes are located at lower levels, structural damage nodes are located at higher levels, and directed edges represent causal relationships and transmission paths.

[0078] Conditional dependency parameters are parameters set between nodes at the same level based on the damage evolution dependency relationship. The damage evolution dependency relationship refers to the mutual influence relationship between nodes at the same level. For example, the state of one structurally damaged node may affect the state of another structurally damaged node. Conditional dependency parameters describe the degree and manner of this mutual influence.

[0079] When constructing a hierarchical transmission network, the node connection relationships are determined based on the set of directed edges, and the nodes are arranged according to the hierarchical structure. Then, conditional dependency parameters are set between nodes at the same level based on the damage evolution dependency relationship. For example, pier displacement affects the stress state of the beam, and conditional dependency parameters are set between the nodes of the two to describe the degree of influence.

[0080] By constructing hierarchical transmission networks and setting condition-dependent parameters, the complex relationship between environmental factors and structural damage can be accurately represented, providing detailed information for Bayesian network reasoning and analysis.

[0081] Step S260: Perform cyclic detection and integration processing on the set of environmental factor nodes, the set of structural damage nodes, and the hierarchical transmission network, remove redundant directed edges that may cause probability propagation conflicts, and generate an initial Bayesian network topology that conforms to the probabilistic inference rules.

[0082] The iterative detection and integration process examines the set of environmental factor nodes, the set of structural damage nodes, and the hierarchical propagation network to identify and remove potentially redundant directed edges that could lead to probability propagation conflicts. Probability propagation conflicts occur in Bayesian networks when redundant directed edges or unreasonable node connections cause contradictory or illogical results in probability propagation.

[0083] Redundant directed edges are those that can be replaced by other paths when representing the causal relationship between environmental factors and structural damage. These edges may cause probability propagation conflicts and affect the reasoning results.

[0084] In one implementation, step S260 may specifically include the following steps S261 to S266:

[0085] Step S261: According to the spatial distribution characteristics of the bridge structure, the nodes in the environmental factor node set are assigned to different spatial areas of the bridge, so that the spatial position of the environmental factor node corresponds to the actual position of action.

[0086] The spatial distribution characteristics of a bridge structure refer to its spatial layout and structural features, including the location and distribution of components such as piers, beams, and foundations. The nodes in the environmental factor node set represent various environmental factors, such as water flow velocity and precipitation. Assigning these nodes to different spatial regions of the bridge ensures that their spatial locations correspond to their actual positions of influence, more accurately representing the impact of environmental factors on different parts of the bridge.

[0087] When assigning nodes, the spatial locations of different parts of the bridge are determined based on the bridge's design drawings and geographic information data. Then, according to the actual range and degree of influence of environmental factors, nodes are assigned to corresponding areas. For example, since water flow mainly affects the bridge piers, the water flow velocity node can be assigned to the area where the piers are located. By assigning environmental factor nodes to different spatial areas of the bridge, the initial Bayesian network topology is made more consistent with reality, improving the network's ability to describe and analyze the bridge's structural safety status in the face of water damage.

[0088] Step S262: Arrange the nodes in the set of structural damage nodes into a time series according to the time sequence of bridge structural damage evolution, so that the order of structural damage nodes corresponds to the stage of damage development.

[0089] The temporal sequence of bridge structural damage evolution refers to the chronological order and stage characteristics of damage development under environmental influences. The nodes in the structural damage node set represent various structural damage states, such as stress-strain and displacement monitoring information. Arranging these nodes into a time series ensures that the node order corresponds to the damage development stages, clearly demonstrating the structural damage development process. When arranging the nodes, the damage development sequence and stage characteristics are determined based on the theory of bridge structural damage evolution and actual monitoring data. For example, when a bridge is impacted by water flow, the instantaneous stress response of the piers first occurs, followed by an increase in cumulative stress, ultimately leading to pier displacement and foundation scour. Based on this process, the nodes related to the instantaneous stress response of the piers are placed first, the nodes related to cumulative stress are placed in the middle, and the nodes related to pier displacement and foundation scour are placed last.

[0090] Step S263: Associate the direct and indirect transmission paths in the hierarchical transmission network with the nodes in the spatial region and time series to construct a spatial-temporal fusion network framework.

[0091] The space-time fusion network framework is a network framework that integrates spatial and temporal information, and can more comprehensively represent the relationship between environmental factors and structural damage. The direct and indirect transmission paths in the hierarchical transmission network represent causal relationships and transmission paths, while the nodes in the spatial region and time series represent the spatial location and temporal sequence of environmental factors and structural damage, respectively.

[0092] When constructing a space-time fusion network framework, direct and indirect transmission paths are associated with nodes in spatial regions and time series. Specifically, directed edges in the paths are connected to environmental factors and structural damage nodes in the spatial region, so that the starting and ending points of the directed edges correspond to actual spatial locations. At the same time, directed edges are associated with nodes in the time series, so that the transmission process conforms to the temporal sequence of structural damage development.

[0093] For example, the direct interaction path between the water flow velocity node and the pier displacement node will have the starting point of the directed edge corresponding to the spatial region where the pier is located, and the ending point will also correspond to this region. Furthermore, the position of the directed edge in the time series can be determined based on the temporal sequence of the pier displacement development.

[0094] Step S264: In the network framework, connect the environmental factor nodes to the corresponding structural damage nodes through directed edges. Directed edges of the direct action path connect nodes at the same time stage, and directed edges of the indirect transmission path connect nodes at different time stages.

[0095] In the space-time fusion network framework, environmental factor nodes and structural damage nodes are arranged according to spatial region and time sequence. The causal relationship between environmental factor nodes and their corresponding structural damage nodes is clarified by connecting them with directed edges.

[0096] A direct action path represents an environmental factor that directly causes structural damage, with directed edges connecting nodes at the same time point. For example, an increase in water flow velocity directly causes pier displacement, and the directed edges connecting the two nodes reflect the impact at the same point in time.

[0097] An indirect transmission path represents how environmental factors indirectly cause structural damage through intermediate physical effects, with directed edges connecting nodes at different time stages. For example, an increase in water flow velocity causes a rise in river water level, which in turn increases the water pressure on the bridge foundation, ultimately leading to foundation settlement. The directed edges of this indirect transmission path reflect the impact at different points in time.

[0098] Step S265: Perform loop detection on the node connection relationships in the network framework to find potential node connection loops that may lead to probabilistic inference loops.

[0099] A node connection loop is a closed loop formed by nodes connected by directed edges in a network framework. A probabilistic inference loop occurs in a Bayesian network when the presence of node connection loops leads to infinite loops or illogical results in probability propagation.

[0100] When constructing the initial Bayesian network topology, the complex node connections may lead to potential node connection loops. These loops can affect the inference results, causing inaccuracies or failure to converge. Therefore, it is necessary to perform loop detection on the node connections to find potential loops.

[0101] When performing cycle detection, graph theory algorithms such as depth-first search (DFS) and topology sorting can be used. DFS traverses network nodes and directed edges to check for cycles. Topology sorting sorts nodes; if it cannot complete the sorting, a cycle exists. By detecting cycles, potential cycles are found, network problems are identified and resolved promptly, ensuring the rationality and reliability of the initial Bayesian network topology.

[0102] Step S266: Remove redundant directed edges in the detected node connection loops, retain directed edges that conform to the time sequence and physical action logic, and generate the initial Bayesian network topology.

[0103] Once a node connection loop is detected in the network framework, redundant directed edges in the loop must be removed. Redundant directed edges are edges that can be replaced by other paths when there is a causal relationship between environmental factors and structural damage.

[0104] Removing redundant directed edges eliminates node connection loops, ensuring the proper functioning of probabilistic reasoning. During removal, directed edges that conform to temporal order and physical action logic are retained. Temporal order refers to the direction of edge propagation conforming to the sequence of structural damage development, while physical action logic refers to the causal relationship represented by the directed edges conforming to the physical process of environmental factors acting on structural damage.

[0105] For example, in a detected loop, there is a directed edge connecting two nodes, but the propagation process does not conform to the time sequence or the causal relationship does not conform to physical logic. This edge is redundant and needs to be removed.

[0106] Step S300: Input environmental impact data and structural response data into the initial Bayesian network topology for parameter learning, adjust the conditional probability distribution of each node in the network, and generate an updated network topology with real-time adjustment capabilities.

[0107] Parameter learning is the process of estimating the conditional probability distribution of each node in an initial Bayesian network topology using known environmental action data and structural response data. The conditional probability distribution is the probability that a node is in a different state given the states of other nodes. For example, the probability that a bridge pier displacement is in a different state given water flow velocity and precipitation.

[0108] Inputting environmental impact data and structural response data into the initial Bayesian network topology aims to adjust the conditional probability distributions of nodes using this data. Through continuous learning and adjustment, the network more accurately reflects the probabilistic relationship between environmental factors and structural damage. Real-time adjustment capability means that the updated network topology can promptly adjust the conditional probability distributions of nodes based on new environmental impact and structural response data to adapt to changes in environmental and structural conditions. For example, when new water flow velocity and precipitation data appear, the conditional probability distributions of pier displacement nodes can be quickly adjusted.

[0109] In one implementation, step S300 may specifically include the following steps S310 to S360:

[0110] Step S310: Perform feature extraction processing on the water flow monitoring information and precipitation monitoring information in the environmental impact data to obtain the environmental feature vector sequence corresponding to each node in the environmental factor node set. The length of the environmental feature vector sequence corresponds to the data acquisition period.

[0111] Feature extraction is the process of extracting characteristic information reflecting environmental factors from water flow monitoring and precipitation monitoring information. An environmental feature vector is a vector containing multiple feature values, each corresponding to a node in the set of environmental factor nodes. An environmental feature vector sequence is a sequence of environmental feature vectors arranged in chronological order, with its length corresponding to the data acquisition period. Water flow monitoring information contains various characteristics of water flow, and precipitation monitoring information contains features of precipitation. Through feature extraction, feature values ​​corresponding to the environmental factor nodes are extracted from this information. For example, if the set of environmental factor nodes includes water flow velocity nodes and precipitation amount nodes, feature extraction extracts water flow velocity feature values ​​from the water flow monitoring information and precipitation amount feature values ​​from the precipitation monitoring information, combining them into an environmental feature vector with a dimension consistent with the number of nodes.

[0112] In one implementation, step S310 may specifically include the following steps S311 to S315:

[0113] Step S311: Segment the water flow monitoring information. Divide the water flow monitoring information into continuous time periods according to the data collection time interval, with each time period corresponding to a data segment.

[0114] The data acquisition time interval is the time difference between two consecutive data acquisitions of water flow monitoring information. Segmentation processing divides the water flow monitoring information into continuous time periods according to the time interval, with each time period corresponding to a data segment.

[0115] Water flow monitoring information is continuously recorded data that includes water flow characteristics at different points in time. By segmenting the data, it is divided into independent segments, which facilitates subsequent feature calculation and analysis.

[0116] For example, if the data collection interval is fixed, the water flow monitoring information can be divided into time periods according to this duration. Each time period corresponds to a data segment, which contains the water flow monitoring data within that time period, such as water flow velocity and flow rate.

[0117] When performing segmented processing, the water flow monitoring information is arranged in chronological order based on the timestamp information of the data acquisition, and then divided into time intervals. This can be achieved using loop structures and conditional statements in programming languages.

[0118] Step S312: Perform feature calculation on the water flow monitoring information for each time period, and extract the intensity feature and change feature of the water flow. The intensity feature is calculated based on the maximum value in the data segment, and the change feature is calculated based on the change amplitude in the data segment.

[0119] Intensity characteristics reflect the magnitude of water flow intensity and are calculated based on the maximum value in a data segment. For example, the maximum water flow velocity within a time period is used as the intensity characteristic of the water flow during that time period. Variation characteristics reflect the variation of water flow and are calculated based on the magnitude of variation in a data segment. For example, the difference between the maximum and minimum water flow velocity within a time period is used as the variation characteristic.

[0120] When performing feature calculations on water flow monitoring information for each time period, the maximum and minimum values ​​are first determined from the data segments. This can be achieved using loop structures and comparison statements in programming languages. For example, for a time period containing multiple water flow velocity data, the data is iterated to determine the maximum and minimum values. Then, the intensity feature is calculated based on the maximum value, and the variation feature is calculated based on the difference between the maximum and minimum values. For instance, after determining the maximum and minimum water flow velocity values ​​within a time period, the intensity and variation features can be obtained.

[0121] Step S313: Perform the same segmentation and feature calculation on the precipitation monitoring information to extract the intensity and duration features of precipitation. The intensity features are calculated based on the cumulative values ​​in the data segments, and the duration features are calculated based on the time length of the data segments.

[0122] The method for segmenting precipitation monitoring information is similar to that for water flow monitoring information. Precipitation monitoring information is divided into continuous time periods according to the data collection time interval, and each time period corresponds to a data segment.

[0123] Precipitation intensity characteristics reflect the magnitude of precipitation and are calculated based on cumulative values ​​within a data segment. For example, the cumulative precipitation over a given time period serves as the intensity characteristic for that period. Duration characteristics reflect the duration of precipitation and are calculated based on the length of a data segment. For example, the length of a given time period serves as the duration characteristic for the precipitation during that period.

[0124] When performing feature calculations on precipitation monitoring information for each time period, the cumulative precipitation is first calculated from the data segments. This can be achieved using loop structures and accumulation statements in programming languages. For example, for a time period containing multiple precipitation data points, the data can be iterated and accumulated to obtain the cumulative precipitation. Then, the intensity feature is calculated based on the cumulative precipitation, and the duration feature is calculated based on the duration of the data segments. For example, once the cumulative precipitation and duration of a time period are determined, the intensity and duration features can be obtained.

[0125] Step S314: Combine the extracted water flow features and precipitation features into an environmental feature vector. The dimension of the environmental feature vector is consistent with the number of nodes in the environmental factor node set, and each dimension corresponds to one environmental factor node.

[0126] An environmental feature vector is a vector containing multiple feature values. It combines extracted water flow and precipitation features to represent the comprehensive characteristics of environmental factors. Its dimensions are consistent with the number of nodes in the environmental factor node set, with each dimension corresponding to one node.

[0127] For example, the set of environmental factor nodes includes nodes for water flow velocity, precipitation, water flow change characteristics, and precipitation duration characteristics, and the environmental feature vector has a dimension of 4. The water flow intensity and change characteristics extracted in step S312, and the precipitation intensity and duration characteristics extracted in step S313, are arranged according to their corresponding nodes to form a 4-dimensional environmental feature vector.

[0128] When combining environmental feature vectors, ensure that each feature value corresponds to the correct environmental factor node. Feature values ​​can be sequentially placed into the corresponding dimensions of the vector according to the order of nodes in the node set.

[0129] Step S315: Arrange the environmental feature vectors in chronological order to generate an environmental feature vector sequence. The order of the vectors in the environmental feature vector sequence corresponds to the chronological order of the data collection.

[0130] An environmental feature vector sequence is a sequence of environmental feature vectors arranged in chronological order, reflecting the changes in environmental factors over time. The order of the vectors in the sequence is consistent with the chronological order of data collection, that is, the vector corresponding to the earliest collected data is listed first, and the vector corresponding to the latest collected data is listed last.

[0131] When generating the environmental feature vector sequence, the environmental feature vectors are sorted according to the timestamp information of the data collection. Sorting algorithms from programming languages, such as quicksort and bubble sort, can be used. For example, if there are multiple environmental feature vectors, each corresponding to data collected at different times, these vectors are arranged in chronological order to form the environmental feature vector sequence.

[0132] Step S320: Perform state identification processing on the stress-strain monitoring information and displacement monitoring information in the structural response data to obtain a structural state identification sequence corresponding to each node in the set of structural damage nodes. The number of states in the structural state identification sequence corresponds to the degree level of structural damage.

[0133] State labeling is the process of converting stress-strain monitoring information and displacement monitoring information into state labels corresponding to each node in the structural damage node set. The structural state label sequence is a sequence containing multiple state labels, with each state label corresponding to a structural damage node.

[0134] The severity level of structural damage is classified according to the degree of damage, such as minor damage, moderate damage, and severe damage. The number of states in the structural state identifier sequence corresponds to the severity level, with each state identifier representing one level.

[0135] When processing the status identification, the numerical ranges of stress, strain, and displacement monitoring information are divided into different intervals, with each interval corresponding to a structural damage level. For example, for bridge pier displacement monitoring information, it is divided into different intervals according to the displacement magnitude, with each interval corresponding to a damage level. Then, based on the monitoring information values ​​for each time period, the corresponding interval is determined, and the corresponding status identifier is added to the structural status identifier sequence.

[0136] Step S330: Input the environmental feature vector sequence and the structural state identifier sequence into the initial Bayesian network topology, so that the environmental feature vector sequence matches the set of environmental factor nodes and the structural state identifier sequence matches the set of structural damage nodes.

[0137] The environmental feature vector sequence and the structural state identifier sequence are input into the initial Bayesian network topology to enable the network to learn the probabilistic relationship between environmental factors and structural damage. The environmental feature vector sequence contains the feature information of environmental factors and corresponds to the set of environmental factor nodes; the structural state identifier sequence contains the state information of structural damage and corresponds to the set of structural damage nodes.

[0138] During the input process, it is essential to ensure that each vector in the environmental feature vector sequence corresponds one-to-one with a node in the environmental factor node set, and each state identifier in the structural state identifier sequence corresponds one-to-one with a node in the structural damage node set. This allows the network to accurately adjust the conditional probability distribution of each node based on the input data.

[0139] For example, the values ​​of each dimension of the first vector in the environmental feature vector sequence correspond to the values ​​of each node in the set of environmental factor nodes. When this vector is input into the network, these values ​​are assigned to the corresponding nodes. Similarly, the first state identifier in the structural state identifier sequence corresponds to the corresponding node in the set of structural damage nodes. When this identifier is input into the network, the corresponding state is set for that node.

[0140] By correctly inputting the environmental feature vector sequence and the structural state identifier sequence into the initial Bayesian network topology, the network can use this data to learn parameters, thereby more accurately reflecting the relationship between environmental factors and structural damage.

[0141] Step S340: Adjust the conditional probability distribution of each node in the initial Bayesian network topology. The conditional probability distribution is updated based on the input environmental feature vector sequence and the structural state identifier sequence. The update process is completed through iterative calculation.

[0142] Adjusting the conditional probability distribution of each node in the initial Bayesian network topology is a crucial step in parameter learning, aiming to enable the network to more accurately describe the probabilistic relationship between environmental factors and structural damage. Updating the conditional probability distribution depends on the input sequence of environmental feature vectors and the sequence of structural state identifiers. Iterative computation can be used during the update process. Iterative computation is a continuously repeating process; each iteration adjusts the conditional probability distribution based on the current input data and the network's existing state, gradually improving the network's fit to the data.

[0143] In one implementation, step S340 may include the following steps S341 to S346:

[0144] Step S341: Initialize the conditional probability distribution of each node in the initial Bayesian network topology. The initial conditional probability distribution of environmental factor nodes is determined based on the statistical values ​​of the environmental feature vector sequence, and the initial conditional probability distribution of structural damage nodes is determined based on the statistical values ​​of the structural state identifier sequence.

[0145] Initializing the conditional probability distribution of each node provides a starting point for subsequent iterative calculations. For environmental factor nodes, their initial conditional probability distribution is determined based on the statistical values ​​of the environmental feature vector sequence. For example, the mean, standard deviation, and other statistics of each feature dimension in the environmental feature vector sequence can be calculated, and the initial probability of the environmental factor node being in different states can be determined based on this.

[0146] For a structurally damaged node, its initial conditional probability distribution is determined based on the statistical values ​​of the structural state identifier sequence. The frequency of different state identifiers appearing in the structural state identifier sequence can be statistically analyzed, and these frequencies can be used as the initial probabilities of the structurally damaged node being in different states.

[0147] For example, in an environmental feature vector sequence, the distribution of the feature values ​​corresponding to a certain environmental factor node across different vectors follows a certain pattern. By statistically analyzing these feature values, the initial conditional probability distribution of that environmental factor node can be determined. Similarly, in a structural state identifier sequence, the frequency of different structural damage state identifiers is counted and used as the initial conditional probability of the structural damage node.

[0148] Step S342: Input the first vector in the environmental feature vector sequence into the initial Bayesian network topology, calculate the conditional probability of each structural damage node under the environmental feature vector, and perform the calculation based on the initial conditional probability distribution.

[0149] Inputting the first vector from the environmental feature vector sequence into the network marks the start of iterative computation. After inputting this vector, the conditional probability of each structurally damaged node being in different states under the current environmental feature vector is calculated based on the initial conditional probability distribution.

[0150] The calculation process is based on the probability propagation mechanism of Bayesian networks. The directed edges in the network represent the causal relationships between nodes. Based on these relationships and the initial conditional probability distribution, starting from the environmental factor nodes, the probability is propagated step by step to the structural damage nodes through probability calculation to obtain the conditional probability of each structural damage node under the environmental feature vector.

[0151] For example, when the first environmental feature vector is input, the network calculates the probability that each structural damage node is in a different state such as slight damage, moderate damage, or severe damage under the environmental feature, based on the directed edge connection relationship between environmental factor nodes and structural damage nodes and the initial conditional probability distribution.

[0152] Step S343: Compare the calculated conditional probabilities with the corresponding states in the structural state identifier sequence, and adjust the conditional probability distribution of the structural damage nodes according to the comparison results. The adjustment magnitude corresponds to the degree of difference in the comparison results.

[0153] Comparing the calculated conditional probabilities of structural damage nodes with the corresponding time points in the structural state identification sequence is to evaluate the consistency between the network calculation results and the actual observation results. If there is a discrepancy, it indicates that the current conditional probability distribution may be inaccurate and needs to be adjusted.

[0154] The adjustment range is determined based on the degree of difference in the comparison results. The greater the difference, the larger the adjustment range; the smaller the difference, the smaller the adjustment range. In this way, the conditional probability distribution of the network is gradually made to better reflect reality.

[0155] For example, if the calculated conditional probability of a structural damage node being in a certain state is inconsistent with the actual state of that node at the corresponding time point in the structural state identification sequence, and the difference is large, then it is necessary to adjust the conditional probability distribution of the structural damage node significantly to make it closer to the actual situation.

[0156] Step S344: Process the subsequent vectors in the environmental feature vector sequence in sequence, repeating the process of calculating conditional probabilities, comparing differences, and adjusting conditional probability distributions until all vectors have been processed.

[0157] Processing subsequent vectors in the environmental feature vector sequence sequentially is a continuous iterative calculation process. For each new vector, steps S342 and S343 are repeated, namely, calculating the conditional probability of the structural damage node under that vector, comparing it with the corresponding state in the structural state identifier sequence, and adjusting the conditional probability distribution according to the difference.

[0158] By continuously repeating this process, the conditional probability distribution of the network will be continuously optimized, gradually reflecting the probabilistic relationship between environmental factors and structural damage more accurately.

[0159] Step S345: Calculate the degree of matching between the adjusted conditional probability distribution and the actual structural state identifier sequence. The degree of matching is comprehensively evaluated based on the conditional probability calculation results of all vectors.

[0160] Calculating the matching degree between the adjusted conditional probability distribution and the actual structural state identifier sequence is to evaluate the effectiveness of the iterative computation. The evaluation of the matching degree is based on a comprehensive consideration of the conditional probability calculation results of all vectors. Various methods can be used to evaluate the matching degree, such as calculating the similarity and error rate between the conditional probability distribution and the actual state identifiers. Through comprehensive evaluation, it can be determined whether the network's conditional probability distribution has sufficiently accurately reflected the relationship between environmental factors and structural damage.

[0161] For example, the similarity between the conditional probability distribution of each structural damage node under all environmental feature vectors and the actual structural state identification sequence is calculated, and these similarities are combined to obtain an overall matching degree index.

[0162] Step S346: Stop adjusting when the matching degree meets the requirements; otherwise, increase the number of iterations and continue adjusting until the conditional probability distribution meets the requirements.

[0163] When the calculated matching degree reaches the preset requirements, it indicates that the conditional probability distribution of the network can reflect the probabilistic relationship between environmental factors and structural damage well, and the iterative adjustment process is stopped at this time.

[0164] If the matching degree does not meet the requirements, it means that the conditional probability distribution of the network needs to be further optimized. At this time, the number of iterations is increased, and the operation of steps S342-S345 is repeated until the conditional probability distribution meets the requirements.

[0165] For example, a high threshold is set for the degree of matching. If the calculated degree of matching is lower than the threshold, iterative adjustments are made to continuously optimize the conditional probability distribution until the degree of matching reaches or exceeds the threshold.

[0166] Step S350: Introduce the time interval between data acquisition time and the current time as a correction parameter. When the time interval between the data sample acquisition time and the current time is less than or equal to the time interval threshold, the contribution of the data sample to the probability estimation is calculated according to the first proportional coefficient. When the time interval between the data sample acquisition time and the current time is greater than the time interval threshold, the contribution of the data sample to the probability estimation is calculated according to the second proportional coefficient. The first proportional coefficient is greater than the second proportional coefficient. The time interval threshold is determined jointly based on the acquisition cycle of environmental impact data and the time scale characteristics of bridge structure damage evolution.

[0167] Introducing the time interval between data acquisition and the current moment as a correction parameter is to account for the impact of data timeliness on probability estimation. Data acquired at different times may have varying importance for the current probability estimation; newer data typically reflects the current environment and structural state better and therefore requires higher weighting.

[0168] The time interval threshold is the boundary for distinguishing the timeliness of data, determined jointly based on the collection cycle of environmental impact data and the timescale characteristics of bridge structural damage evolution. When the time interval between the data sample collection time and the current time is less than or equal to the threshold, it indicates that the data is relatively new, and its contribution to probability estimation is calculated according to the first proportional coefficient. When the time interval is greater than the threshold, it indicates that the data is relatively old, and its contribution is calculated according to the second proportional coefficient, with the first proportional coefficient being greater than the second proportional coefficient. For example, for recently collected data, the time interval between its collection time and the current time is less than the time interval threshold, and it is given a higher weight in the first proportional coefficient when performing probability estimation, making its impact on probability estimation greater; while for earlier collected data, the time interval is greater than the threshold, and it is given a lower weight in the second proportional coefficient, reducing its impact on probability estimation.

[0169] Step S360: When new environmental action data and structural response data arrive, the conditional probability distribution is re-estimated based on the sliding window mechanism. The window size is negatively correlated with the rate of change of environmental action, and an updated network topology is generated.

[0170] The sliding window mechanism selects a fixed-length window on the data sequence and only considers the data within the window for probability estimation. The window size is negatively correlated with the rate of change of environmental factors; that is, when the rate of change of environmental factors is fast, the window size decreases to include more recent high-frequency changes; when the rate of change of environmental factors is slow, the window size increases to include more historical data.

[0171] In one implementation, step S360 may specifically include the following steps S361 to S367:

[0172] Step S361: Perform timestamp sequence detection on the continuously arriving new environmental action data and structural response data. When the time interval between the timestamp of the new data and the timestamp of the most recent parameter update reaches the time interval threshold, start the parameter update. The timestamp interval is calculated by accumulating the time difference between the new data and the historical data.

[0173] Timestamp sequence detection of newly arrived data is used to determine when parameter updates need to be initiated. The timestamp interval is a key indicator for determining whether to update parameters. The parameter update process is initiated when the interval between the timestamp of the new data and the timestamp of the most recent parameter update reaches a pre-set threshold. The timestamp interval is calculated by accumulating the time difference between new data and historical data, thus accurately grasping the timeliness and changes in the data.

[0174] For example, the system continuously records the timestamp of each parameter update. When new environmental impact data and structural response data arrive, its timestamp is compared with the timestamp of the most recent update. If the accumulated time difference reaches a time interval threshold, it indicates that the data has been updated and changed sufficiently, and the network parameters need to be re-estimated and adjusted to ensure that the network can accurately reflect the current environmental and structural state.

[0175] Step S362: Perform feature extraction and state identification processing on the newly received environmental action data and structural response data to generate a new environmental feature vector sequence and a new structural state identification sequence.

[0176] Feature extraction and status labeling of newly received data are performed to transform the raw data into a format suitable for network processing. For new environmental impact data, feature extraction is performed using a method similar to that in step S310. Specifically, the new water flow monitoring information and precipitation monitoring information are segmented and feature-calculated to extract the intensity and variation characteristics of water flow, and the intensity and duration characteristics of precipitation, etc. These are then combined to form a new environmental feature vector, which is then arranged in chronological order to generate a new environmental feature vector sequence.

[0177] For new structural response data, the state identification process is also performed according to the method in step S320. The stress-strain monitoring information and displacement monitoring information are divided into states. Based on the correspondence between the monitored values ​​and the degree of structural damage, a new structural state identification sequence corresponding to each node in the set of structural damage nodes is obtained. The number of states in this sequence corresponds to the degree level of structural damage.

[0178] For example, new water flow monitoring information, after being segmented and feature-calculated, yields new water flow intensity and variation characteristics. These are combined with new precipitation characteristics to form a new environmental feature vector, which is then arranged in chronological order. New stress-strain and displacement monitoring information, after being processed by state labeling, forms a new structural state label sequence, providing accurate data input for subsequent parameter updates.

[0179] Step S363: Calculate the rate of change of environmental impact data, which is characterized by the feature difference between the new environmental feature vector sequence and the historical environmental feature vector sequence within the most recent sliding window. The feature difference is obtained by weighted summation of the change magnitude of each dimension of the vector, and the weight is positively correlated with the correlation strength value of the corresponding environmental factor node.

[0180] The rate of change of environmental impact data is calculated to determine the size of the sliding window. The rate of change is characterized by comparing the feature differences between the new environmental feature vector sequence and the historical environmental feature vector sequence within the most recent sliding window.

[0181] Feature dissimilarity is calculated by weighted summation of the magnitudes of change in each dimension of the vectors. First, the magnitudes of change in each dimension of the corresponding vectors in the new environment feature vector sequence and the historical environment feature vector sequence are calculated. Then, weights are assigned to each dimension based on the correlation strength value of the corresponding environmental factor nodes; the higher the correlation strength value, the higher the weight. Finally, the magnitudes of change in each dimension are multiplied by their respective weights and summed to obtain the feature dissimilarity.

[0182] For example, for the water flow velocity and precipitation dimensions in the environmental feature vector, the magnitude of change in these two dimensions is calculated for the corresponding vectors in the new and historical sequences, respectively. If the correlation between the water flow velocity node and the structural damage node is strong, then the magnitude of change in the water flow velocity dimension will be given a higher weight when calculating the feature difference, which can more accurately reflect the actual changes in the environmental impact data.

[0183] Step S364: Dynamically adjust the size of the sliding window according to the rate of change of the environment. The greater the rate of change, the smaller the time span of the sliding window, so that the window contains more recent high-frequency change data. The smaller the rate of change, the larger the time span of the sliding window. The window size adjustment range is constrained by the time interval threshold.

[0184] The size of the sliding window is dynamically adjusted based on the calculated rate of change of environmental effects. When the rate of change of environmental effects is large, it indicates that the environment is changing rapidly, and more recent data needs to be considered. Therefore, the time span of the sliding window is reduced to include more recent high-frequency change data, so as to capture the impact of environmental changes on structural damage more promptly.

[0185] When the rate of change of environmental influences is small, it indicates that the environment is relatively stable. In this case, the time span of the sliding window can be expanded to include more historical data. This allows for a comprehensive consideration of long-term environmental and structural changes, making the probability estimation more accurate.

[0186] Meanwhile, the window size adjustment range is constrained by the time interval threshold to ensure that the window size does not exceed a reasonable range, thus ensuring a balance between the timeliness and accuracy of the data.

[0187] For example, if the rate of change of environmental factors suddenly increases, the time span of the sliding window will be reduced accordingly, and only data within the most recent period will be selected for analysis; if the rate of change decreases, the time span of the window will be expanded, but will not exceed the range allowed by the time interval threshold.

[0188] Step S365: Within the adjusted sliding window, the new environmental feature vector sequence and the new structural state identifier sequence are weighted and fused with the historical sequence within the window. Data samples with time intervals less than or equal to the time interval threshold within the window are weighted by a first proportional coefficient, and data samples with time intervals greater than the time interval threshold are weighted by a second proportional coefficient to generate a window fusion sequence.

[0189] After adjusting the sliding window size, the new environmental feature vector sequence and the new structural state identifier sequence are weighted and fused with the historical sequence within the window. For data samples with time intervals less than or equal to a time interval threshold within the window, due to their high timeliness, a first proportional coefficient is applied for weighting, giving them a higher weight. For data samples with time intervals greater than the time interval threshold, due to their relatively weak timeliness, a second proportional coefficient is applied for weighting, giving them a lower weight. This weighted fusion process rationally combines new and historical data to generate a windowed fusion sequence. This sequence contains information that comprehensively considers both new and old data, better reflecting changes in the environment and structural state.

[0190] Step S366: Re-estimate the conditional probability distribution of each node based on the window fusion sequence. The re-estimate process adopts the same iterative calculation logic as the initial parameter learning, and the number of iterations is dynamically determined according to the amount of data in the window fusion sequence.

[0191] Re-estimating the conditional probability distribution of each node based on the windowed fusion sequence is to update the network to adapt to new environments and structural states. The re-estimation process employs the same iterative computation logic as the initial parameter learning: starting with the initial conditional probability distribution, data from the windowed fusion sequence is input sequentially, the conditional probabilities of structurally damaged nodes are calculated, compared with the actual state, and the conditional probability distribution is adjusted accordingly. This process iterates until certain conditions are met. The number of iterations is dynamically determined based on the amount of data in the windowed fusion sequence. If the data volume is large, more iterations may be needed to fully learn the information in the data and make the conditional probability distribution more accurate; if the data volume is small, the number of iterations can be reduced accordingly to improve computational efficiency.

[0192] Step S367: Calculate the difference between the re-estimated conditional probability distribution and the distribution before the update. The difference is calculated by the average change of the conditional probability values ​​of each structural damage node. When the difference exceeds the difference threshold dynamically determined by the rate of change of environmental effects, the re-estimated conditional probability distribution is integrated into the network topology to generate an updated network topology. The difference threshold is positively correlated with the rate of change of environmental effects.

[0193] Calculating the difference between the re-estimated conditional probability distribution and the original distribution is to determine whether the re-estimated result has changed sufficiently and whether the network topology needs to be updated. The difference is calculated by the average change in the conditional probability values ​​of each structurally damaged node; that is, calculating the change in the conditional probability value of each structurally damaged node before and after re-estimated, and then averaging the results.

[0194] The difference threshold is dynamically determined by the rate of change of environmental influences and is positively correlated with it. When the rate of change of environmental influences is large, the difference threshold will also increase accordingly, because the environmental and structural states change rapidly, allowing for larger changes in the conditional probability distribution; when the rate of change of environmental influences is small, the difference threshold will decrease, requiring relatively small changes in the conditional probability distribution.

[0195] When the difference exceeds the difference threshold, it indicates that the re-estimated conditional probability distribution has changed sufficiently to better reflect the current environment and structural state. At this point, the re-estimated conditional probability distribution is integrated into the network topology to generate an updated network topology.

[0196] For example, if the conditional probability values ​​of each structural damage node change significantly after re-estimation, and the calculated difference exceeds the difference threshold determined by the rate of change of the current environmental influence, then the new conditional probability distribution is applied to the network to update the network topology, enabling the network to perform subsequent state reasoning more accurately.

[0197] Step S400: Based on the updated network topology, integrate the real-time collected environmental action data and structural response data, perform network state reasoning processing, and obtain state reasoning results that reflect the current safety level of the bridge structure.

[0198] Updating the network topology reflects the latest probabilistic relationship between environmental factors and structural damage. By integrating real-time collected environmental action data and structural response data for network state inference processing, the current safety level of the bridge structure can be accurately assessed.

[0199] Real-time collected environmental action data and structural response data reflect the current actual state of the bridge. By inputting these data into the updated network topology and using the network's probabilistic reasoning mechanism, the state probability distribution of each structural damage node can be calculated, thereby obtaining the state reasoning result that reflects the current safety level of the bridge structure.

[0200] In one implementation, step S400 may specifically include the following steps S410 to S470:

[0201] Step S410: For the real-time collected environmental impact data, identify abnormal data in the water flow monitoring information and precipitation monitoring information. The abnormal data is judged based on the degree of deviation of the data value from the normal range, and the abnormal data is corrected.

[0202] Real-time collected environmental data may contain anomalies, which could be caused by monitoring equipment malfunctions, external interference, or other reasons. Identifying anomalies is crucial to ensuring the accuracy of data input to the network.

[0203] By setting a normal range, the data values ​​from water flow monitoring and precipitation monitoring are compared with the normal range to determine whether the data is abnormal. The normal range can be determined based on statistical analysis of historical data, industry standards, or experience.

[0204] For identified abnormal data, correction processing is performed. Correction methods can include interpolation and smoothing filtering. Interpolation can estimate the value of abnormal data points based on adjacent normal data points; smoothing filtering can remove abnormal fluctuations by smoothing the data.

[0205] Step S420: Extract features from the corrected environmental impact data to generate a real-time environmental feature vector. The dimension of the real-time environmental feature vector is consistent with the number of environmental factor nodes in the updated network topology. The feature extraction method is the same as that in the parameter learning stage.

[0206] Feature extraction of the corrected environmental impact data is performed to transform the data into a feature vector format suitable for network processing. The dimension of the generated real-time environmental feature vector is consistent with the number of environmental factor nodes in the updated network topology, ensuring that the feature vector can accurately match the environmental factor nodes in the network.

[0207] The feature extraction method is the same as the parameter learning stage, which involves segmenting and calculating features from the water flow monitoring information and precipitation monitoring information. For water flow monitoring information, the intensity and variation features of the water flow are extracted; for precipitation monitoring information, the intensity and duration features of precipitation are extracted. These features are then combined into a real-time environmental feature vector.

[0208] For example, the corrected water flow monitoring information is segmented, and the water flow intensity characteristics and variation characteristics of each time period are calculated; the same processing is performed on the precipitation monitoring information to extract precipitation characteristics. The extracted water flow characteristics and precipitation characteristics are combined into a real-time environmental feature vector according to the correspondence between environmental factor nodes.

[0209] Step S430: Input the real-time environmental feature vector to update the network topology, set the state of environmental factor nodes, and infer the state of environmental factor nodes that are not directly monitored through existing environmental feature vectors.

[0210] Inputting real-time environmental feature vectors into updating the network topology is to initiate the network's state inference process. Based on the values ​​of the real-time environmental feature vectors, the states of environmental factor nodes in the network topology are set and updated.

[0211] For some environmental factor nodes that are not directly monitored, their state can be inferred from existing environmental feature vectors. By utilizing the causal relationships and probability distributions between nodes in the network, the state of unmonitored nodes can be estimated based on existing environmental feature information. For example, in updating the network topology, there might be an environmental factor node representing a parameter of water flow, but this parameter lacks direct monitoring data. The state of this unmonitored node can be inferred based on existing monitoring data such as water flow velocity and flow rate, combined with the relationships and probability distributions between nodes in the network.

[0212] Step S440: Calculate the state probability distribution of each structurally damaged node in the updated network topology. The state probability distribution is calculated based on the updated conditional probability distribution and the real-time environmental feature vector. The calculation process is achieved through directed edge propagation in the network.

[0213] Calculating the state probability distribution of each structurally damaged node is the core step in network state inference. Based on the updated conditional probability distribution and real-time environmental feature vectors, the calculation process is implemented through directed edges in the network. The directed edges in the network represent the causal relationships between nodes. Starting from environmental factor nodes, the probability is gradually propagated to structurally damaged nodes according to the real-time environmental feature vectors and conditional probability distributions, obtaining the probability of each structurally damaged node being in different states.

[0214] In one implementation, step S440 may specifically include the following steps S441 to S446:

[0215] Step S441: Convert the updated network topology into a loop-free connection structure. Simplify the network structure by merging parent nodes with common child nodes. The simplification process preserves the conditional dependencies between nodes.

[0216] Converting the updated network topology to an acyclic connection structure facilitates probability calculations and inference. Merging parent nodes with shared child nodes simplifies the network structure, reduces computational complexity, and preserves the conditional dependencies between nodes, ensuring the accuracy of probability calculations.

[0217] For example, in updating network topology, if multiple environmental factor nodes point to the same structural damage node, these environmental factor nodes with common child nodes are merged to form a simpler network structure, but the conditional dependencies between them and the structural damage node are still preserved.

[0218] Step S442: Determine the message passing order of each node in the network structure. The message passing order is determined based on the hierarchical relationship of the nodes, starting from the environmental factor node and proceeding to the structural damage node.

[0219] Determining the message passing order among nodes ensures the orderly execution of probability calculations. Based on the hierarchical relationship between nodes, messages are passed from environmental factor nodes to structural damage nodes in a predetermined order. Environmental factor nodes are at lower levels of the network, while structural damage nodes are at higher levels. Message passing starts from the environmental factor nodes and proceeds step-by-step along the directed edges to the structural damage nodes, allowing for the gradual calculation of the state probability distribution of each structural damage node. For example, in the simplified network structure, the message passing order of the environmental factor nodes is first determined, and then messages are passed sequentially to intermediate nodes and structural damage nodes, performing calculations in an orderly manner according to the hierarchical relationship.

[0220] Step S443: Initialize the message value of each node. The message value is set based on the conditional probability distribution of the node. The message value of the environmental factor node is determined according to the real-time environmental feature vector.

[0221] Initializing the message values ​​of each node is the initial step in message passing. Message values ​​are set based on the node's conditional probability distribution. For environmental factor nodes, their message values ​​are determined according to the real-time environmental feature vector. The message value of an environmental factor node reflects the node's probability information under the current environmental state; it is determined based on the feature values ​​in the real-time environmental feature vector and the node's conditional probability distribution. For other nodes, message values ​​are initialized according to their conditional probability distribution. For example, for an environmental factor node, its initial message value is determined based on the value of the corresponding feature in the real-time environmental feature vector, combined with the node's conditional probability distribution. For structurally damaged nodes, message values ​​are initialized based on their conditional probability distribution in updating the network topology.

[0222] Step S444: Calculate the messages that each node transmits to its neighboring nodes according to the message transmission order. The message calculation is based on the node's own conditional probability distribution and the received input messages.

[0223] Following a predetermined message passing order, each node passes messages to its neighboring nodes. Message computation is based on the node's own conditional probability distribution and the received input messages.

[0224] Each node calculates and transmits new messages to its neighbors based on its own conditional probability distribution and the messages it receives from neighboring nodes. In this way, messages are gradually transmitted through the network, continuously updating the probability information of each node.

[0225] For example, after receiving a message from a real-time environmental feature vector, an environmental factor node calculates and transmits a message to the structural damage node connected to it based on its own conditional probability distribution and the message; after receiving a message from a neighboring node, a structural damage node calculates and transmits a new message to other neighboring nodes based on its own conditional probability distribution.

[0226] Step S445: The node receiving the message updates its own message value. The updated message value is used to pass on to subsequent nodes until all structurally damaged nodes have received the relevant messages.

[0227] The receiving node updates its own message value based on the received message. The updated message value is then used to pass on to subsequent nodes, continuing the message passing process. This process is repeated until all structurally damaged nodes have received the relevant messages. At this point, the message value of each structurally damaged node contains information passed from environmental factor nodes, reflecting the probability information of each structurally damaged node being in different states under the current environmental conditions. For example, after receiving a message from a neighboring node, a structurally damaged node updates its own message value based on the message and its own conditional probability distribution, and then passes the updated message value to other neighboring nodes until all structurally damaged nodes have completed message reception and updating.

[0228] Step S446: Calculate the state probability distribution of the structurally damaged node based on the message value received. Each state in the state probability distribution corresponds to a probability value, and the sum of the probability values ​​is 1, reflecting the possibility that the structurally damaged node is in different states.

[0229] Based on the message values ​​received by the structurally damaged node, its state probability distribution is calculated. Each state in the state probability distribution corresponds to a probability value, and the sum of these probability values ​​is 1, reflecting the likelihood of the structurally damaged node being in different states. The state probability distribution is obtained by normalizing the message values ​​received by the structurally damaged node. Normalization ensures that the sum of the probability values ​​of all states is 1, meeting the basic requirements of a probability distribution.

[0230] Step S450: Process the real-time acquired structural response data, and convert the stress-strain monitoring information and displacement monitoring information into status identifiers of structural damage nodes. The status identifiers are determined based on the correspondence between the monitoring values ​​and the degree of structural damage.

[0231] Processing the real-time acquired structural response data aims to transform the actual monitoring data into a form that can be combined with network inference results. Stress-strain monitoring information and displacement monitoring information are converted into status identifiers for structural damage nodes, and these status identifiers are determined based on the correspondence between monitoring values ​​and the degree of structural damage.

[0232] The monitoring value range corresponding to different structural damage levels is preset. When the monitoring values ​​of the real-time collected stress-strain monitoring information and displacement monitoring information fall within a certain range, the corresponding structural damage node is marked as the corresponding state.

[0233] For example, for bridge pier displacement monitoring information, a displacement less than a certain value is defined as a minor damage state, a displacement within a certain range as a moderate damage state, and a displacement greater than a certain value as a severe damage state. Once the real-time collected bridge pier displacement monitoring values ​​are determined, the structural damage nodes corresponding to the bridge pier displacement are marked with the corresponding state based on this correspondence.

[0234] Step S460: Combine the state identifier of the structural damage node with the calculated state probability distribution to generate the comprehensive state probability of the structural damage node. The comprehensive state probability takes into account the direct identifier of the monitoring data and the probability distribution of network inference.

[0235] By combining the state identifiers of structurally damaged nodes with the calculated state probability distributions, a comprehensive state probability is generated. This aims to more accurately assess the state of structurally damaged nodes by comprehensively utilizing information from monitoring data and network inference. The comprehensive state probability considers both the direct identifiers from monitoring data and the probability distributions from network inference. The direct identifiers from monitoring data reflect the actual structural state, while the probability distributions from network inference consider environmental factors and causal relationships between nodes.

[0236] In one implementation, step S460 may specifically include the following steps S461 to S466:

[0237] Step S461: Convert the state identifier of the structural damage node into a probability form. Set the state probability corresponding to the state identifier as the first probability value, and set the other state probabilities as the second probability value. The first probability value is greater than the second probability value, and the sum of the probability values ​​is 1.

[0238] Converting the state identifiers of structural damage nodes into probabilistic form unifies the processing of direct identifiers from monitoring data with the probability distribution of network inference. The state probability corresponding to each state identifier is set as the first probability value, and the probabilities of other states are set as second probability values, with the first probability value being greater than the second probability value, while ensuring that the sum of all state probability values ​​is 1.

[0239] For example, if the state of a structural damage node is identified as minor damage, the probability of the minor damage state is set to a higher first probability value, the probabilities of the moderate damage and severe damage states are set to lower second probability values, and the sum of the probability values ​​of these three states is 1.

[0240] Step S462: Combine the converted state identifier probability with the calculated state probability distribution. The combination process is achieved by weighted summation of the corresponding state values ​​of the two probabilities.

[0241] The converted state identifier probabilities are combined with the calculated state probability distribution using a weighted summation method. A certain weight is assigned to the corresponding state values ​​of the two probabilities, and then the probability values ​​of the corresponding states are multiplied by their respective weights and summed to obtain the combined probability value. For example, for a certain state of a structurally damaged node, the probability value of that state in the state identifier probability is multiplied by one weight, and the probability value of that state in the calculated state probability distribution is multiplied by another weight. The two results are then summed to obtain the combined probability value of that state.

[0242] Step S463: Obtain the preset weighted summation coefficients, wherein the weight coefficients of the state identifier probability are determined based on the reliability of the monitoring data, and the weight coefficients of the calculated state probability distribution are determined based on the credibility of network inference.

[0243] Obtaining preset weighted summation coefficients is to reasonably allocate the proportion of monitoring data and network inference results in the comprehensive state probability calculation. The weight coefficients for state identification probabilities are determined based on the reliability of the monitoring data; if the accuracy and stability of the monitoring data are high, the weight coefficients will be larger; conversely, they will be smaller.

[0244] The weight coefficients of the calculated state probability distribution are determined based on the reliability of network inference. If the network training and update process is reliable and accurately reflects the relationship between the environment and structural state, the weight coefficients will be larger; otherwise, they will be smaller. For example, if the monitoring equipment has high accuracy and the monitoring data is highly reliable, then the weight coefficients of the state identification probability can be set larger; if the updated network topology has undergone sufficient training and validation and the network inference is highly reliable, then the weight coefficients of the calculated state probability distribution can be set larger.

[0245] Step S464: Normalize the weighted sum of probability values ​​so that the sum of the probabilities of each state in the overall state probability is 1. The normalization process is based on the sum of the overall probability values.

[0246] Normalizing the weighted sum of probability values ​​ensures that the overall state probability conforms to the basic requirement of a probability distribution, i.e., the sum of the probability values ​​for all states is 1. Normalization is performed based on the sum of the overall probability values. The weighted sum of the probability values ​​for each state is divided by the sum of the overall probability values ​​to obtain the normalized probability value. For example, after weighted summation, the probability values ​​of the structurally damaged node in different states are obtained. These probability values ​​are added together to get the overall probability value. Then, the probability value of each state is divided by this sum to obtain the normalized overall state probability value, and the sum of these values ​​is 1.

[0247] Step S465: Check the rationality of the overall state probability. The rationality is judged based on the distribution characteristics of each state probability value. The overall state probability with abnormal distribution is recalculated by adjusting the weight coefficient.

[0248] Checking the rationality of the overall state probability is to ensure that it accurately reflects the actual state of the structurally damaged nodes. The distribution characteristics of each state probability value are used for judgment. If the distribution of probability values ​​is not reasonable—for example, if the probability value of a certain state is too high or too low, inconsistent with reality—then the overall state probability is considered abnormal. For abnormally distributed overall state probabilities, the weighting coefficients of the weighted sum are adjusted and recalculated. Different combinations of weighting coefficients are tried until a reasonable overall state probability distribution is obtained.

[0249] For example, if the probability value of a structural damage node being in a severely damaged state in the overall state probability is too high, but actual monitoring data shows that the structural state is relatively good, then this overall state probability is considered unreasonable. The overall state probability is recalculated by adjusting the weighting coefficients of the state identifier probability and the calculated state probability distribution.

[0250] Step S466: Determine the state with the highest probability value in the comprehensive state probability as the current main state of the structural damage node. The probability value of the main state is used as the comprehensive state probability value of the structural damage node for subsequent calculation of the overall safety level index.

[0251] The state with the highest probability value in the comprehensive state probability is determined as the current primary state of the structural damage node. This primary state reflects the most likely state in which the structural damage node is located. The probability value of the primary state is used as the comprehensive state probability value of the structural damage node, which is then used for subsequent calculations of the overall safety level index.

[0252] Step S470: Based on the comprehensive state probability of each structural damage node, calculate the overall safety level index of the bridge structure. The overall safety level index corresponds to the weighted sum of the comprehensive state probabilities of each structural damage node, which is the state reasoning result.

[0253] Based on the comprehensive state probability of each structural damage node, the overall safety level index of the bridge structure is calculated. The overall safety level index is a comprehensive indicator that reflects the current overall safety status of the bridge structure.

[0254] The overall safety level index corresponds to the weighted sum of the comprehensive state probabilities of each structural damage node. A weight is assigned to the comprehensive state probability value of each structural damage node, determined based on the node's importance within the bridge structure. The overall safety level index of the bridge structure is obtained by multiplying the comprehensive state probability values ​​of each structural damage node by their respective weights and then summing them. This index represents the state inference result.

[0255] For example, different weights are assigned to different structural damage nodes, such as piers and beams, based on their importance in the bridge structure. The overall state probability value of each structural damage node is multiplied by its corresponding weight, and then all results are summed to obtain the overall safety level index of the bridge structure, which reflects the current safety level of the bridge.

[0256] Step S500: Based on the state reasoning results, generate a safety state assessment result that includes structural damage risk ranking and identification of key risk factors.

[0257] The state reasoning results reflect the current safety level of the bridge structure. Based on these results, a safety state assessment is generated, including a ranking of structural damage risks and identification of key risk factors.

[0258] Structural damage risk ranking involves prioritizing the risk of each structural damage node based on its overall state probability and its importance within the bridge structure. Nodes with higher risk are ranked higher, while those with lower risk are ranked lower.

[0259] Key risk factor identification involves determining the environmental factors that significantly impact nodes at high risk of structural damage. By analyzing the directed edge connections and probability distributions between nodes in the updated network topology, it is determined which environmental factors have the most significant impact on nodes at high risk of structural damage.

[0260] In one implementation, step S500 may specifically include the following steps S510 to S570:

[0261] Step S510: Extract the comprehensive state probability value of each structural damage node in the state reasoning result. The comprehensive state probability value reflects the probability that the structural damage node is in the current state.

[0262] The comprehensive state probability value of each structural damage node is extracted from the state inference results. These probability values ​​reflect the likelihood of each structural damage node being in its current state. The comprehensive state probability value is determined in step S466, taking into account both monitoring data and network inference results. For example, for different structural damage nodes such as pier displacement and beam stress, their respective comprehensive state probability values ​​are extracted from the state inference results. These values ​​can intuitively reflect the current damage probability of each structural damage node.

[0263] Step S520: Calculate the risk value of each structural damage node. The risk value is calculated based on the comprehensive state probability value and the importance of the structural damage node. The importance is determined according to the components of the bridge structure.

[0264] The risk value of each structural damage node is calculated, taking into account both the overall state probability value and the importance of the structural damage node. The importance of a structural damage node is determined based on its role and position in the bridge structure. For example, bridge piers are key components supporting the bridge and are therefore relatively important; while some secondary auxiliary structures are relatively less important.

[0265] The risk value of each structural damage node is obtained by multiplying the overall state probability value by the importance of the damaged node. A higher risk value indicates a greater potential threat to bridge safety. For example, for a structural damage node involving pier displacement, the overall state probability value is high, and since the pier is crucial to the bridge structure, the calculated risk value for this node will be significant. Conversely, for a damage node in a less important ancillary structure, even if its overall state probability value is high, its risk value will be relatively low due to its lower importance.

[0266] Step S530: Sort the risk values ​​of the structural damage nodes in descending order to generate a structural damage risk ranking. The ranking result reflects the order of risk of different structural damage nodes. The ranking process is based on the magnitude of the risk value.

[0267] The risk values ​​of each structural damage node are sorted in descending order to generate a structural damage risk ranking. The ranking results clearly reflect the order of risk of different structural damage nodes, facilitating the assessment and management of the safety status of bridge structures.

[0268] The sorting process is based on the magnitude of the risk value. A sorting algorithm is used to arrange the risk values ​​from largest to smallest, and the corresponding structural damage nodes are also sorted accordingly.

[0269] For example, by sorting the risk values ​​of structural damage nodes such as pier displacement, beam cracks, and bearing damage using a sorting algorithm, a structural damage risk ranking from high to low can be obtained. This makes it easy to see which structural damage nodes pose the greatest threat to bridge safety.

[0270] Step S540: Identify environmental factor nodes that have a significant impact on high-risk structural damage nodes that rank high in the risk ranking. The degree of impact is determined based on the directed edge connections in the updated network topology.

[0271] Identifying environmental factor nodes that significantly impact high-risk structural damage nodes in the risk ranking is crucial for determining the key factors leading to bridge structural damage. The degree of impact is determined based on the directed edge connections in the updated network topology, where directed edges represent causal relationships between nodes.

[0272] In updating the network topology, if an environmental factor node is connected to a high-risk structural damage node through a directed edge, and the connection is strong, it indicates that the environmental factor node has a significant impact on the structural damage node.

[0273] For example, in risk ranking, the pier displacement node is ranked high, indicating a high-risk structural damage node. If, during network topology updates, it is found that the water flow velocity node is connected to the pier displacement node via directed edges with strong connections, then the water flow velocity node can be considered an environmental factor node with a significant impact on pier displacement.

[0274] Step S550: Calculate the degree of influence of each environmental factor node on the high-risk structural damage node. The degree of influence is comprehensively evaluated by the number of directed edges and the connection strength between the environmental factor node and the structural damage node.

[0275] The impact of each environmental factor node on high-risk structural damage nodes is calculated, taking into account both the number of directed edges and the connection strength between environmental factor nodes and structural damage nodes. A larger number of directed edges indicates a closer connection between environmental factor nodes and structural damage nodes; greater connection strength indicates a more significant impact of environmental factors on structural damage.

[0276] For example, by analyzing the updated network topology, the number of directed edges between a certain environmental factor node and a high-risk structural damage node can be counted, and the connection strength of each directed edge can be evaluated. By comprehensively considering the number of directed edges and the connection strength, the degree of influence of the environmental factor node on the high-risk structural damage node can be calculated.

[0277] Step S560: Identify the environmental factor nodes with a high degree of influence as key risk factors, and establish a correspondence between key risk factors and high-risk structural damage nodes.

[0278] Environmental factors with a high degree of impact were identified as key risk factors, which are the main causes of damage at high-risk structural damage nodes. A correspondence was established between key risk factors and high-risk structural damage nodes, clarifying which environmental factors caused each high-risk structural damage node.

[0279] For example, calculations and assessments revealed that environmental factors such as water flow velocity and precipitation have a significant impact on high-risk structural damage nodes such as pier displacement and beam cracks. These environmental factors were identified as key risk factors, and a correspondence between them and high-risk structural damage nodes was established, such as water flow velocity corresponding to pier displacement and precipitation corresponding to beam cracks.

[0280] Step S570: Integrate the structural damage risk ranking and the correspondence between key risk factors to generate a safety status assessment result that includes risk order, risk factors, and risk values. The format of the assessment result should conform to the output requirements of the bridge safety monitoring system.

[0281] By integrating the structural damage risk ranking with the correspondence between key risk factors, a safety status assessment result is generated. The assessment result includes information such as risk ranking, risk factors, and risk values, comprehensively reflecting the safety status of the bridge structure.

[0282] The evaluation results are formatted in accordance with the output requirements of the bridge safety monitoring system, facilitating system display, storage, and analysis. The evaluation results can be presented in the form of reports, charts, etc., intuitively demonstrating the safety status of the bridge structure.

[0283] For example, the structural damage risk can be ranked and presented in a list, outlining the risk order and risk value for each structural damage node. Simultaneously, key risk factors can be listed for each high-risk structural damage node. Integrating this information generates a safety status assessment result that meets the output requirements of a bridge safety monitoring system, providing a strong basis for bridge safety management and maintenance.

[0284] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention can all be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solutions of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set thresholds based on historical data, experience or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select activation functions, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0285] Figure 2 This is a schematic diagram of the structural composition of a bridge structure water damage safety status monitoring device provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the bridge structure water damage safety status monitoring device 200 includes:

[0286] The data acquisition module 210 is used to acquire environmental action data and structural response data of the area where the bridge is located. The environmental action data includes water flow monitoring information reflecting hydrodynamic conditions and precipitation monitoring information reflecting meteorological change characteristics. The structural response data includes stress and strain monitoring information characterizing the mechanical behavior of bridge components and displacement monitoring information characterizing the overall deformation characteristics.

[0287] The network construction module 220 is used to combine historical water damage event records and bridge structural damage evolution patterns to construct an initial Bayesian network topology to describe the relationship between environmental factors and structural damage. The initial Bayesian network topology includes a set of directed edges that reflect the transmission path of environmental effects and a node hierarchy that represents the state of structural damage.

[0288] The network update module 230 is used to input the environmental action data and structural response data into the initial Bayesian network topology for parameter learning, adjust the conditional probability distribution of each node in the network, and generate an updated network topology with real-time adjustment capability.

[0289] The state reasoning module 240 is used to perform network state reasoning processing based on the updated network topology, integrate real-time collected environmental action data and structural response data, and obtain a state reasoning result that reflects the current safety level of the bridge structure.

[0290] The status assessment module 250 is used to generate a safety status assessment result that includes a ranking of structural damage risks and identification of key risk factors based on the status reasoning result.

[0291] The description of the above device embodiments is similar to that of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the device embodiments of the present invention, please refer to the description of the method embodiments of the present invention for understanding.

[0292] Figure 3 A hardware entity diagram of a computer system provided as an embodiment of the present invention, such as... Figure 3 As shown, the hardware entity of the computer system 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

Claims

1. A method for monitoring the safety status of bridge structures damaged by water based on Bayesian networks, characterized in that, The method includes: The environmental impact data and structural response data of the bridge area are obtained. The environmental impact data includes water flow monitoring information reflecting hydrodynamic conditions and precipitation monitoring information reflecting meteorological change characteristics. The structural response data includes stress and strain monitoring information characterizing the mechanical behavior of bridge components and displacement monitoring information characterizing the overall deformation characteristics. Based on historical records of water damage events and the evolution of bridge structural damage, an initial Bayesian network topology is constructed to describe the relationship between environmental factors and structural damage. The initial Bayesian network topology includes a set of directed edges that reflect the transmission path of environmental effects and a hierarchical relationship of nodes that represent the state of structural damage. The environmental impact data and structural response data are input into the initial Bayesian network topology for parameter learning, adjusting the conditional probability distribution of each node in the network to generate an updated network topology with real-time adjustment capabilities. Specifically, this includes: performing feature extraction processing on the water flow monitoring information and precipitation monitoring information in the environmental impact data to obtain an environmental feature vector sequence corresponding to each node in the environmental factor node set, the length of which corresponds to the data acquisition period; and performing state identification processing on the stress-strain monitoring information and displacement monitoring information in the structural response data to obtain a structural state identification sequence corresponding to each node in the structural damage node set, the number of states in the structural state identification sequence corresponding to the structural damage... The severity level of the injury corresponds to the following: The environmental feature vector sequence and the structural state identifier sequence are input into the initial Bayesian network topology, so that the environmental feature vector sequence matches the set of environmental factor nodes, and the structural state identifier sequence matches the set of structural damage nodes; the conditional probability distribution of each node in the initial Bayesian network topology is initialized, the initial conditional probability distribution of environmental factor nodes is determined based on the statistical values ​​of the environmental feature vector sequence, and the initial conditional probability distribution of structural damage nodes is determined based on the statistical values ​​of the structural state identifier sequence; the first vector in the environmental feature vector sequence is input into the initial Bayesian network topology, and the conditional probability of each structural damage node under that environmental feature vector is calculated, based on the initial conditional probability distribution; the calculated... The calculated conditional probabilities are compared with the corresponding states in the structural state identifier sequence. Based on the comparison results, the conditional probability distribution of the structural damage nodes is adjusted, with the adjustment magnitude corresponding to the degree of difference in the comparison results. Subsequent vectors in the environmental feature vector sequence are processed sequentially, repeating the process of calculating conditional probabilities, comparing differences, and adjusting the conditional probability distribution until all vectors are processed. The degree of matching between the adjusted conditional probability distribution and the actual structural state identifier sequence is calculated, with the degree of matching comprehensively evaluated based on the conditional probability calculation results of all vectors. Adjustment stops when the degree of matching meets the requirements; otherwise, the number of iterations is increased to continue adjusting until the conditional probability distribution meets the requirements. The time interval between data acquisition time and the current moment is introduced as... The parameters are adjusted as follows: when the time interval between the data sample acquisition time and the current time is less than or equal to the time interval threshold, the contribution of the data sample to the probability estimation is calculated according to the first proportional coefficient; when the time interval between the data sample acquisition time and the current time is greater than the time interval threshold, the contribution of the data sample to the probability estimation is calculated according to the second proportional coefficient. The first proportional coefficient is greater than the second proportional coefficient. The time interval threshold is jointly determined based on the acquisition cycle of environmental action data and the time scale characteristics of bridge structural damage evolution. When new environmental action data and structural response data arrive, the conditional probability distribution is re-estimated based on the sliding window mechanism. The window size is negatively correlated with the rate of change of environmental action, and an updated network topology is generated. Based on the updated network topology, the real-time collected environmental action data and structural response data are integrated to perform network state reasoning processing, and a state reasoning result reflecting the current safety level of the bridge structure is obtained. Based on the state reasoning results, a safety state assessment result is generated, which includes a ranking of structural damage risks and the identification of key risk factors.

2. The method according to claim 1, characterized in that, The initial Bayesian network topology for describing the relationship between environmental factors and structural damage is constructed by combining historical records of flood damage events and the evolution patterns of bridge structural damage, including: Spatiotemporal correlation mining was performed on the correlation characteristics between environmental factors and structural damage in historical flood damage records to obtain a set of causal correlation descriptions of environmental factors and structural damage at different time stages and spatial locations. The correlation strength between environmental factors and structural damage in the causal correlation description set is filtered to generate a correlation strength sequence between environmental factors and structural damage. The values ​​in the correlation strength sequence are related to the frequency and duration of the co-occurrence of environmental factors and structural damage. The node types of the initial Bayesian network topology are divided according to the correlation strength sequence to obtain the set of environmental factor nodes and the set of structural damage nodes. The set of environmental factor nodes corresponds to various influencing factors in the environmental action data, and the set of structural damage nodes corresponds to various damage states in the structural response data. By analyzing the physical process of environmental factors acting on structural damage, the direction of the directed edges between the set of environmental factor nodes and the set of structural damage nodes is determined, and a set of directed edges containing direct action paths and indirect transmission paths is generated. Based on the set of directed edges, a hierarchical transmission network from environmental factor nodes to structural damage nodes is constructed. Nodes at the same level are configured with conditional dependency parameters according to the dependency relationship of damage evolution. The set of environmental factor nodes, the set of structural damage nodes, and the hierarchical transmission network are cyclically detected and integrated to remove redundant directed edges that may cause probability propagation conflicts, thereby generating an initial Bayesian network topology that conforms to the probabilistic inference rules.

3. The method according to claim 2, characterized in that, The step of filtering the correlation strength between environmental factors and structural damage in the causal correlation description set to generate a correlation strength sequence between environmental factors and structural damage includes: The environmental action parameters in the causal association description set are subjected to multi-feature decomposition processing to extract the impact and scour features from the water flow monitoring information, and the intensity and duration features from the precipitation monitoring information. The structural damage indicators in the causal association description set are divided into multiple dimensions, and the stress-strain monitoring information is decomposed into instantaneous response indicators and cumulative effect indicators, and the displacement monitoring information is decomposed into local deformation indicators and overall deformation indicators. The correlation strength values ​​between various environmental action parameters and structural damage indicators are calculated based on a multi-scale sliding window. The length of the sliding window is adjusted according to the duration characteristics of the environmental action. The calculated correlation strength values ​​are aligned over time to ensure that the correlation strength values ​​of different environmental action parameters and structural damage indicators are on the same time axis. A correlation strength sequence is generated based on the correlation strength values ​​after time series alignment. Each element in the correlation strength sequence corresponds to the degree of correlation between environmental factors and structural damage within a specific time window. The association strength sequences are subjected to hierarchical clustering to merge sequence segments with similar association patterns, generating a set of association strength sequences representing different association patterns.

4. The method according to claim 3, characterized in that, The process of analyzing the physical process of environmental factors acting on structural damage to determine the direction of directed edges between the set of environmental factor nodes and the set of structural damage nodes, and generating a set of directed edges containing direct action paths and indirect transmission paths, includes: The environmental action parameters corresponding to each node in the set of environmental factor nodes are analyzed for action time to determine the start time and duration of each environmental factor acting on the bridge structure. The occurrence time of structural damage indicators corresponding to each node in the set of structural damage nodes is analyzed to determine the initial time and development time of each structural damage. By comparing the time difference between the time of environmental factors' action and the time of structural damage, the direction of the causal relationship between environmental factors and structural damage can be determined. For environmental factor nodes and structural damage nodes that have a direct causal relationship, direct edges are established to form a direct action path; For environmental factor nodes and structural damage nodes with indirect causal relationships, an indirectly connected directed edge chain is established by introducing intermediate transmission nodes to form an indirect transmission path. The intermediate transmission nodes characterize the intermediate physical effects of environmental factors on structural damage. Path redundancy detection is performed on directed edges in direct action paths and indirect propagation paths. Redundant directed edges that can be replaced by other paths are removed, and a set of directed edges is generated. The process of cyclically detecting and integrating the set of environmental factor nodes, the set of structural damage nodes, and the hierarchical transmission network to generate an initial Bayesian network topology that conforms to probabilistic inference rules includes: Based on the spatial distribution characteristics of the bridge structure, the nodes in the environmental factor node set are assigned to different spatial areas of the bridge, so that the spatial location of the environmental factor node corresponds to its actual position. The nodes in the set of structural damage nodes are arranged into a time series according to the time sequence of bridge structural damage evolution, so that the order of structural damage nodes corresponds to the stage of damage development. By associating the direct and indirect transmission paths in the hierarchical transmission network with nodes in spatial regions and time series, a spatial-temporal fusion network framework is constructed. In the network framework, environmental factor nodes are connected to corresponding structural damage nodes through directed edges. Directed edges of direct action paths connect nodes at the same time stage, while directed edges of indirect transmission paths connect nodes at different time stages. Perform loop detection on the node connection relationships in the network framework to find potential node connection loops that may lead to probabilistic inference loops. Remove redundant directed edges from the detected node connection loops, retain directed edges that conform to the time sequence and physical action logic, and generate the initial Bayesian network topology.

5. The method according to claim 1, characterized in that, The process of extracting features from the water flow monitoring information and precipitation monitoring information in the environmental impact data yields an environmental feature vector sequence corresponding to each node in the environmental factor node set, including: The water flow monitoring information is segmented and divided into continuous time periods according to the data collection time interval, with each time period corresponding to a data segment. Feature calculations are performed on the water flow monitoring information for each time period to extract the intensity and variation features of the water flow. The intensity features are calculated based on the maximum value in the data segment, and the variation features are calculated based on the variation amplitude in the data segment. The precipitation monitoring information is processed in the same way as the previous segmentation and feature calculation to extract the intensity and duration features of precipitation. The intensity features are calculated based on the cumulative values ​​in the data segments, and the duration features are calculated based on the time length of the data segments. The extracted water flow features and precipitation features are combined into an environmental feature vector. The dimension of the environmental feature vector is consistent with the number of nodes in the set of environmental factor nodes, and each dimension corresponds to one environmental factor node. The environmental feature vectors are arranged in chronological order to generate an environmental feature vector sequence, and the order of the vectors in the environmental feature vector sequence corresponds to the chronological order of data collection.

6. The method according to claim 1, characterized in that, When new environmental action data and structural response data arrive, the conditional probability distribution is re-estimated based on a sliding window mechanism. The window size is negatively correlated with the rate of change of environmental action, generating an updated network topology, including: Timestamp sequence detection is performed on continuously arriving new environmental action data and structural response data. When the time interval between the timestamp of the new data and the timestamp of the most recent parameter update reaches the time interval threshold, parameter update is initiated. The timestamp interval is calculated by accumulating the time difference between the new data and the historical data. The newly received environmental action data and structural response data are processed for feature extraction and state identification to generate a new environmental feature vector sequence and a new structural state identification sequence. The rate of change of environmental impact data is calculated by the feature difference between the new environmental feature vector sequence and the historical environmental feature vector sequence within the most recent sliding window. The feature difference is obtained by weighted summation of the change magnitude of each dimension of the vector, and the weights are positively correlated with the correlation strength value of the corresponding environmental factor node. The sliding window size is dynamically adjusted according to the rate of change of the environment. The greater the rate of change, the shorter the time span of the sliding window, so that the window contains more recent high-frequency change data. The smaller the rate of change, the longer the time span of the sliding window. The window size adjustment range is constrained by the time interval threshold. Within the adjusted sliding window, the new environmental feature vector sequence and the new structural state identifier sequence are weighted and fused with the historical sequence within the window. Data samples with time intervals less than or equal to the time interval threshold are weighted by a first proportional coefficient, and data samples with time intervals greater than the time interval threshold are weighted by a second proportional coefficient to generate a window fusion sequence. The conditional probability distribution of each node is re-estimated based on the window fusion sequence. The re-estimate process adopts the same iterative calculation logic as the initial parameter learning, and the number of iterations is dynamically determined according to the amount of data in the window fusion sequence. The difference between the re-estimated conditional probability distribution and the original distribution is calculated. The difference is calculated by the average change of the conditional probability values ​​of each structural damage node. When the difference exceeds the difference threshold dynamically determined by the rate of change of environmental effects, the re-estimated conditional probability distribution is integrated into the network topology to generate an updated network topology. The difference threshold is positively correlated with the rate of change of environmental effects.

7. A bridge structure water damage safety status monitoring device, characterized in that, The device includes: The data acquisition module is used to acquire environmental action data and structural response data of the area where the bridge is located. The environmental action data includes water flow monitoring information reflecting hydrodynamic conditions and precipitation monitoring information reflecting meteorological change characteristics. The structural response data includes stress and strain monitoring information characterizing the mechanical behavior of bridge components and displacement monitoring information characterizing the overall deformation characteristics. The network construction module is used to combine historical records of water damage events and the evolution of bridge structural damage to construct an initial Bayesian network topology to describe the relationship between environmental factors and structural damage. The initial Bayesian network topology includes a set of directed edges that reflect the transmission path of environmental effects and a hierarchical relationship of nodes that represent the state of structural damage. The network update module is used to input the environmental impact data and structural response data into the initial Bayesian network topology for parameter learning, adjust the conditional probability distribution of each node in the network, and generate an updated network topology with real-time adjustment capabilities. Specifically, it includes: performing feature extraction processing on the water flow monitoring information and precipitation monitoring information in the environmental impact data to obtain an environmental feature vector sequence corresponding to each node in the environmental factor node set, the length of which corresponds to the data acquisition period; and performing state identification processing on the stress-strain monitoring information and displacement monitoring information in the structural response data to obtain a structural state identification sequence corresponding to each node in the structural damage node set, the state of which is specified in the structural state identification sequence. The quantity corresponds to the degree of structural damage; the environmental feature vector sequence and the structural state identifier sequence are input into the initial Bayesian network topology, so that the environmental feature vector sequence matches the set of environmental factor nodes, and the structural state identifier sequence matches the set of structural damage nodes; the conditional probability distribution of each node in the initial Bayesian network topology is initialized, the initial conditional probability distribution of environmental factor nodes is determined based on the statistical values ​​of the environmental feature vector sequence, and the initial conditional probability distribution of structural damage nodes is determined based on the statistical values ​​of the structural state identifier sequence; the first vector in the environmental feature vector sequence is input into the initial Bayesian network topology, the conditional probability of each structural damage node under this environmental feature vector is calculated, and the conditional probability of each node is calculated based on the initial conditional probability distribution. The process involves comparing the calculated conditional probabilities with the corresponding states in the structural state identifier sequence, adjusting the conditional probability distribution of the structural damage nodes based on the comparison results, with the adjustment magnitude corresponding to the degree of difference in the comparison results; processing subsequent vectors in the environmental feature vector sequence sequentially, repeating the process of calculating conditional probabilities, comparing differences, and adjusting conditional probability distributions until all vectors are processed; calculating the matching degree between the adjusted conditional probability distribution and the actual structural state identifier sequence, with the matching degree being comprehensively evaluated based on the conditional probability calculation results of all vectors; stopping the adjustment when the matching degree meets the requirements, otherwise increasing the number of iterations to continue adjusting until the conditional probability distribution meets the requirements; and introducing the data acquisition time and the time interval between the current moments. As a correction parameter, when the time interval between the data sample acquisition time and the current time is less than or equal to the time interval threshold, the contribution of the data sample to the probability estimation is calculated according to the first proportional coefficient. When the time interval between the data sample acquisition time and the current time is greater than the time interval threshold, the contribution of the data sample to the probability estimation is calculated according to the second proportional coefficient. The first proportional coefficient is greater than the second proportional coefficient. The time interval threshold is jointly determined based on the acquisition cycle of environmental action data and the time scale characteristics of bridge structural damage evolution. When new environmental action data and structural response data arrive, the conditional probability distribution is re-estimated based on the sliding window mechanism. The window size is negatively correlated with the rate of change of environmental action, and an updated network topology is generated. The state reasoning module is used to perform network state reasoning processing based on the updated network topology, integrate real-time collected environmental action data and structural response data, and obtain a state reasoning result that reflects the current safety level of the bridge structure. The status assessment module is used to generate a safety status assessment result that includes a ranking of structural damage risks and identification of key risk factors based on the status reasoning results.

8. A computer system comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 6.

Images