Earthquake-resistant performance grading method for railway bridge structure in near-fault strong earthquake area

Abstract

The application provides a near-fault strong earthquake area railway bridge structure seismic performance grading method, which comprises the following steps: collecting real-time state data of a bridge in a near-fault earthquake area railway; inputting the real-time state data into a dynamic failure network model to obtain risk prediction information; the dynamic failure network model is a pre-constructed weighted directed graph model, and the nodes in the dynamic failure network model represent failure modes of the bridge; inputting the real-time state data and the risk prediction information into a pre-trained deep reinforcement learning intelligent agent to generate control instructions; grading the seismic performance of the bridge based on the risk prediction information, and sending the control instructions to a distributed executor cluster deployed on the bridge to enable the distributed executor cluster to receive and execute the corresponding control instructions to perform graded control on the bridge structure. The possible risk level of the structure is more accurately predicted, the adaptive control strategy is realized, and the problem of lag of traditional pre-defined rule control is reduced.

Description

Technical Field

[0001] This application relates to the field of bridge seismic technology, and more specifically, to a method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones. Background Technology

[0002] Existing methods for assessing the seismic resistance and performance classification of railway bridges under near-fault strong earthquakes are typically static risk assessment methods. These methods only perform offline analysis during the design phase and cannot dynamically adjust based on the real-time response of the structure during a strong earthquake. Furthermore, control strategies are often based on preset fixed rules, making it difficult to cope with changes in the structure and ground motion during an earthquake. Summary of the Invention

[0003] The purpose of this application is to provide a method, program product, electronic device and storage medium for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones, in order to improve the above-mentioned problems.

[0004] In a first aspect, embodiments of this application provide a method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones, comprising: collecting real-time state data of bridges in railways in near-fault earthquake zones; the real-time state data reflecting at least one of the bridge's displacement, strain value, and acceleration; inputting the real-time state data into a dynamic failure network model to obtain risk prediction information; the dynamic failure network model is a pre-constructed weighted directed graph model, where nodes represent bridge failure modes; directed edges between nodes represent the correlation between failure modes; and the weights of directed edges represent the probability information of the failure mode to which the excitation is directed; inputting the real-time state data and risk prediction information into a pre-trained deep reinforcement learning agent to generate control commands; classifying the bridge's seismic performance based on the risk prediction information; and sending the control commands to a distributed actuator cluster deployed on the bridge, so that the distributed actuator cluster receives and executes the corresponding control commands to perform graded control of the bridge structure.

[0005] In the aforementioned implementation process, by collecting structural response data in real time and using a dynamic failure network model for risk simulation, the potential damage paths and risk levels of the structure can be predicted more accurately. By automatically generating optimized control commands through a pre-trained deep reinforcement learning agent, intelligent and adaptive control strategies are achieved, reducing the lag problem of traditional predefined rule control. Through a clear performance grading mechanism, different levels of defense strategies can be matched and activated based on real-time risk prediction results, directing the distributed actuator cluster to work collaboratively.

[0006] Optionally, in this embodiment of the application, real-time status data is input into the dynamic failure network model to obtain risk prediction information, including: calculating the real-time risk value of each node and the weight of each directed edge in the dynamic failure network model based on the real-time status data, and updating the dynamic failure network model to generate an updated dynamic failure network model; using the updated dynamic failure network model to perform risk extrapolation and generate risk prediction information; the risk prediction information is used to represent the probability of occurrence of each failure mode, the failure propagation path, and the expected occurrence time in the future period.

[0007] In the aforementioned implementation process, the performance grading method enables more accurate and forward-looking risk assessment. By dynamically calculating node risk and edge weights, the model can accurately reflect the actual response of the structure in real time, improving the accuracy and timeliness of risk perception. Using the updated model for extrapolation, it is possible to systematically predict potential future failure chains and their timelines, achieving a leap from current situation awareness to future early warning.

[0008] Optionally, in the embodiments of this application, the failure modes represented by the nodes in the dynamic failure network model include at least one of track slab misalignment, support slippage, main beam displacement exceeding limits, pier bending cracking, pier shear failure, and beam collapse.

[0009] In the aforementioned implementation process, by clearly defining the failure modes represented by the nodes in the dynamic failure network model, including key damage patterns such as track slab misalignment and support slippage, the risk identification and analysis objectives of the performance grading method are clearly defined. The modeling approach based on specific failure modes facilitates the subsequent transformation of abstract risks into concrete seismic performance levels, improving the targeted nature of structural safety protection and overall management efficiency.

[0010] Optionally, in the embodiments of this application, the directed edges between nodes represent the association between failure modes. The association between failure modes refers to the causal triggering relationship between the preceding failure mode and the subsequent failure mode. The preceding failure mode refers to the failure mode corresponding to the starting node of the directed edge; the subsequent failure mode refers to the failure mode corresponding to the ending node of the directed edge.

[0011] In the above implementation process, directed edges between nodes are defined to represent the causal relationship between preceding failure modes and subsequent failure modes. The dynamic failure network model can more realistically simulate the inherent logic and chain reaction process of bridge damage evolution under strong earthquakes. This transforms the model from a collection of isolated risk points into a system capable of describing how risks are transmitted and evolved. Risk analysis based on this model can not only identify which failure modes currently pose a high risk, but also reveal where the risks originate, along which path they are most likely to spread, and what serious consequences they may ultimately cause. This provides crucial forward-looking information for subsequent intelligent decision-making and hierarchical control.

[0012] Optionally, in this embodiment of the application, the probability information includes the excitation conditional probability and the propagation delay; the probability information is dynamically updated based on real-time state data.

[0013] In the aforementioned implementation process, the excitation conditional probability and propagation delay can be dynamically updated based on real-time state data. This dynamic update mechanism allows the dynamic failure network model to reduce the limitations of static analysis and capture in real time the true impact of changes in seismic characteristics and accumulated structural damage on the strength of the causal relationship between failures. The failure propagation paths and predicted times derived by the dynamic failure network model will be closer to the actual situation, improving the accuracy and timeliness of risk situation prediction.

[0014] Optionally, in this embodiment of the application, before inputting real-time state data and risk prediction information into a pre-trained deep reinforcement learning agent to generate control commands, the method further includes: constructing an observation space, action space, and reward function for the agent in a digital twin simulation environment of the bridge; the observation space includes real-time state data and the state of the dynamic failure network model, the action space is used to define a set of control commands for the distributed actuator cluster, and the reward function aims to minimize global structural risk and / or bridge component damage; the digital twin environment is stimulated using near-fault seismic waves, enabling the agent to be trained through a reinforcement learning algorithm to obtain a pre-trained deep reinforcement learning agent; wherein, the agent has a pre-installed policy neural network for calculating and outputting control commands based on the current observation state.

[0015] In the aforementioned implementation process, a core with highly adaptable and intelligent decision-making capabilities was obtained by pre-training a deep reinforcement learning agent in a digital twin simulation environment. The agent underwent training on massive amounts of earthquake scenarios in the simulation, mastering strategies for generating efficient control commands when faced with complex, nonlinear real-time state data and risk prediction information. This effectively reduces the rigidity and inability to cope with uncertainty inherent in traditional predefined rule control strategies, thereby significantly improving the command efficiency and collaborative control effect of distributed actuator clusters in real earthquakes. This makes the entire seismic graded control process more intelligent and precise, improving the overall response efficiency and reliability from risk perception to proactive intervention.

[0016] Optionally, in this embodiment of the application, classifying the seismic performance of a bridge based on risk prediction information includes: calculating at least one risk quantification index based on the risk prediction information; the risk quantification index is used to characterize the severity of the risk of the bridge; the risk quantification index is compared with risk thresholds of different levels to determine the seismic performance level of the bridge; wherein, different seismic performance levels correspond to different control priorities.

[0017] In the aforementioned implementation process, by quantifying risk indicators and comparing them with preset risk thresholds, automated classification of bridge risk status was achieved. The relatively complex risk prediction information obtained in the preceding steps was transformed into a clear and explicit seismic performance level decision. Different levels correspond to differentiated control priorities, meaning that the invocation of the distributed actuator cluster is no longer a single mode, but rather matched according to the severity of the risk. This allows the system to operate in a low-power monitoring state when the risk is low, saving energy and reducing equipment wear; when the risk increases, it can automatically and promptly switch to a higher-level defense state, allocating more control resources for precise intervention.

[0018] Optionally, in embodiments of this application, the distributed actuator cluster includes at least one of a shape memory alloy limiting device, a self-healing concrete unit, and a magnetorheological damper.

[0019] In the aforementioned implementation process, a distributed actuator cluster comprising shape memory alloy limiting devices, self-healing concrete units, and magnetorheological dampers is configured. The magnetorheological dampers provide rapid and adjustable vibration control; the shape memory alloy devices provide crucial rigid backup constraints; and the self-healing concrete units enable active repair of material damage. This combination allows the system to flexibly select the most suitable actuators or actuator combinations to implement graded control based on risk prediction information and the determined seismic performance level.

[0020] Optionally, in this embodiment of the application, the method further includes: after the distributed actuator cluster executes the corresponding control command, collecting real-time feedback data of the bridge; transmitting the real-time feedback data to the deep reinforcement learning agent, and optimizing the deep reinforcement learning agent using the real-time feedback data.

[0021] In the above implementation process, by collecting real-time feedback data after executing control and using it to optimize the deep reinforcement learning agent, and continuously fine-tuning and improving its decision-making strategy based on the actual control effect, the agent can increasingly accurately understand the behavioral characteristics of a specific bridge during an earthquake, thereby improving the accuracy and effectiveness of subsequent control commands.

[0022] Optionally, in this embodiment of the application, collecting real-time status data of bridges in railways near fault seismic zones includes: collecting real-time status data of bridges in railways near fault seismic zones through a sensor network deployed on the bridges; the sensor network includes at least one of displacement sensors, strain sensors, and acceleration sensors.

[0023] In the aforementioned implementation process, a sensor network containing various types of sensors is deployed to collect real-time status data, providing comprehensive raw information input for subsequent decision-making in this performance grading method. Displacement, strain, and acceleration data, from different dimensions such as geometric deformation, material internal forces, and kinetic inertia, collectively describe the complete mechanical state of the bridge under seismic loading. This comprehensive data foundation enables more accurate decision-making.

[0024] Secondly, this application also provides a seismic performance grading device for railway bridge structures in near-fault strong earthquake zones, comprising: a data acquisition module for acquiring real-time status data of bridges in railways in near-fault earthquake zones; the real-time status data reflects at least one of the bridge's displacement, strain value, and acceleration; a risk prediction module for inputting the real-time status data into a dynamic failure network model to obtain risk prediction information; the dynamic failure network model is a pre-constructed weighted directed graph model, where nodes represent bridge failure modes; directed edges between nodes represent the correlation between failure modes; and the weights of the directed edges represent the probability information of the failure mode to which the excitation is directed; an instruction generation module for inputting the real-time status data and risk prediction information into a pre-trained deep reinforcement learning agent to generate control instructions; and an execution module for grading the bridge's seismic performance based on the risk prediction information and sending the control instructions to a distributed actuator cluster deployed on the bridge, so that the distributed actuator cluster receives and executes the corresponding control instructions to perform graded control of the bridge structure.

[0025] Thirdly, embodiments of this application also provide a computer program product, including computer program instructions, which are executed by a processor to perform the method provided in the first aspect or any implementation thereof.

[0026] Fourthly, embodiments of this application also provide an electronic device, including: a processor and a memory, the memory storing computer program instructions, which are executed by the processor to perform the method provided in the first aspect or any implementation thereof.

[0027] Fifthly, embodiments of this application also provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, perform the method provided in the first aspect or any implementation thereof.

[0028] The method, program product, electronic equipment, and storage medium for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones provided in this application are as follows: Attached Figure Description To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart illustrating a method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones, provided as an embodiment of this application; Figure 2 A schematic diagram of the seismic performance grading device for railway bridge structures in near-fault strong earthquake zones provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0030] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this application.

[0032] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0033] Please see Figure 1 The illustrated flowchart presents a method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones according to an embodiment of this application. This method can be applied to electronic devices, which may include physical devices such as servers, PCs, tablets, or smartphones, or virtual devices such as virtual machines or containers. The electronic device can be a single device, a combination of multiple devices, or a cluster of numerous devices. The method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones may include: Step S110: Collect real-time status data of bridges in railways near the fault seismic zone; the real-time status data reflects at least one of the bridge's displacement, strain value, and acceleration.

[0034] Step S120: Input real-time status data into the dynamic failure network model to obtain risk prediction information; the dynamic failure network model is a pre-constructed weighted directed graph model, where nodes represent bridge failure modes; directed edges between nodes represent the correlation between failure modes; and the weight of a directed edge represents the probability information of triggering the failure mode it points to.

[0035] Step S130: Input real-time state data and risk prediction information into the pre-trained deep reinforcement learning agent to generate control commands.

[0036] Step S140: Based on the risk prediction information, classify the seismic performance of the bridge and send control commands to the distributed actuator cluster deployed on the bridge so that the distributed actuator cluster receives and executes the corresponding control commands to perform graded control of the bridge structure.

[0037] In step S110, real-time status data refers to physical quantities continuously measured by a sensor network deployed at key parts of the bridge. These quantities may include one or more of displacement, strain, and acceleration. Displacement data can be monitored using high-precision GPS or wire-type displacement gauges installed at locations such as the ends of the main beams and the tops of the piers to capture the relative displacement between the beams and the piers. Strain values ​​are obtained using strain gauges or fiber optic sensors attached to key sections such as the plastic hinge zones of the piers and the mid-span of the main beams, reflecting the degree of deformation of the internal materials of the components. Acceleration data is collected by triaxial accelerometers arranged at various levels of the bridge (such as the bridge deck and pier tops) to analyze the vibration response characteristics of the structure. All sensors transmit data sampled at high frequencies (e.g., 100Hz) synchronously in real time to the central data processing unit via wired or wireless transmission networks, forming a complete data stream describing the instantaneous mechanical state of the bridge.

[0038] In step S120, the dynamic failure network model is a weighted directed graph model, which serves as the computational tool for risk situation assessment in this classification method. In this dynamic failure network model, nodes are predefined as potential failure modes determined through engineering analysis, such as "support slippage," "pier bending and cracking," and "beam collapse." Directed edges connect these nodes, explicitly representing the correlation between one failure mode and another, i.e., the causal propagation direction.

[0039] Each directed edge is assigned a weight, which specifically includes two elements: the trigger conditional probability and the propagation delay. The trigger conditional probability refers to the likelihood that a subsequent failure will be triggered given that a predecessor failure has occurred. The propagation delay refers to the approximate time interval between the occurrence of a predecessor failure and the occurrence of a subsequent failure.

[0040] For example, step S120 can be implemented as follows: First, the real-time status data passed in step S110 is matched with the failure states represented by each node, and the "activation" risk value of each node is calculated and updated. Based on the latest node status, the weight parameters of each directed edge are dynamically adjusted so that the dynamic failure network model reflects the impact of the current seismic excitation in real time. Finally, based on the updated network, a complex network propagation algorithm is used to perform deduction and obtain risk prediction information. The risk prediction information includes the probability of occurrence of each failure mode in the future, multiple key failure propagation chains, and their estimated development timelines.

[0041] In step S130, the pre-trained deep reinforcement learning agent is a software module that has undergone extensive offline training and possesses autonomous decision-making capabilities. The pre-training process is completed in a digital twin simulation environment of the bridge: in this virtual environment, the agent's observation space, action space, and reward function are defined; the observation space is the input, including the simulated real-time data stream from sensors and the state of the dynamic failure network; the action space is the output, which is the set of instructions for all controllable actuators, such as "adjust the output of damper A to the preset value"; the reward function aims to minimize the global risk integral of the network and the damage to key components.

[0042] By inputting thousands of near-fault seismic wave records, the agent learns through repeated trial and error in simulation, ultimately enabling its internal policy neural network to master which action combination to output under what observation conditions to obtain the maximum long-term reward. In actual deployment, the agent fuses the real-time state data from step S110 with the risk prediction information from step S120 into a complete observation state vector, which is then input into the pre-trained policy network. The network directly generates control commands after forward propagation. Control commands are specific, executable instruction sets, such as specifying which actuators to activate or adjust to specified parameters.

[0043] In step S140, firstly, the risk is classified based on the risk prediction information: according to the risk prediction information, such as the global risk value and the probability of critical failure, the bridge's current seismic performance level is automatically determined by comparing it with preset thresholds for different levels, such as "conventional monitoring level," "active intervention level," or "comprehensive defense level." Different levels correspond to different defense intensities and strategic priorities.

[0044] A tiered control approach is adopted: control commands are issued to a distributed actuator cluster. The distributed actuator cluster is a collection of physical actuation devices installed on the bridge, including at least one of magnetorheological dampers, shape memory alloy restraint devices, and self-healing concrete units. During execution, based on the determined performance level, the corresponding subset of actuators is driven or its operating mode is adjusted. For example, at the active intervention level, the main command is to increase the damping force of the magnetorheological dampers to suppress vibration; while at the comprehensive defense level, the magnetorheological dampers are simultaneously commanded to operate at full capacity, the shape memory alloy devices are triggered to rigidify to prevent beam collapse, and the self-healing concrete units in designated areas are activated. The actuators act immediately upon receiving the commands, thereby achieving tiered control of the structural dynamic response and damage development, matched to the risk level.

[0045] As one implementation method, the hierarchical information can also be used to adjust the control priority. For example, if the hierarchical information represents a bridge with low seismic performance, it means that the more urgent the execution action of the bridge, the higher the priority of the control command corresponding to the bridge can be processed.

[0046] In the implementation of the above embodiments: by collecting structural response data in real time and using a dynamic failure network model for risk simulation, the potential damage paths and risk levels of the structure can be predicted more accurately. By automatically generating optimized control commands through a pre-trained deep reinforcement learning agent, intelligent and adaptive control strategies are achieved, reducing the lag problem of traditional predefined rule control. Through a clear performance grading mechanism, different levels of defense strategies can be matched and activated based on real-time risk prediction results, directing the distributed actuator cluster to work collaboratively.

[0047] In this application embodiment, "strong earthquake zone" refers to an area where the peak ground acceleration (PGA) is not less than 0.20g, according to the current national "China Seismic Ground Motion Parameter Zoning Map" (GB18306) or other equivalent engineering seismic fortification standards.

[0048] Optionally, in this embodiment of the application, real-time status data is input into the dynamic failure network model to obtain risk prediction information, including: Based on real-time state data, the real-time risk value of each node and the weight of each directed edge in the dynamic failure network model are calculated, and the dynamic failure network model is updated to generate the updated dynamic failure network model.

[0049] The system matches and analyzes the incoming real-time status data with the trigger threshold or characteristic state of the failure mode represented by each node. For example, it compares the real-time monitored strain value at the bottom of the pier with the material yield strain threshold corresponding to the pier bending crack node, and calculates the real-time risk value of the node at the current moment through a preset membership function or probability model. The real-time risk value can be a value between 0 and 1, representing the probability that the failure is about to occur or is occurring.

[0050] The weights of each directed edge are then calculated and updated. These weights are not fixed; they are dynamically adjusted based on the real-time status of connected nodes and the overall structural behavior reflected in the real-time data. For example, when abnormally severe structural vibration is detected, the excitation conditional probability of the directed edge from the main beam displacement exceeding the limit to the beam falling may be increased by the algorithm, and the propagation delay may be shortened. Weight adjustments can be implemented based on pre-established data-driven models or empirical physical formulas. Finally, using the calculated new node risk values ​​and edge weights, the dynamic failure network model is updated, replacing the old parameters, thus generating an updated dynamic failure network model that reflects the latest, dynamic risk correlation state of the structure under seismic loading.

[0051] Risk prediction information is generated by using the updated dynamic failure network model. The risk prediction information is used to represent the probability of occurrence of each failure mode, the failure propagation path and the expected occurrence time in the future period.

[0052] Risk simulation refers to the process of simulating how failure propagates gradually over time within a model, given a network structure and current parameters. Risk prediction information is the final output of the simulation calculation, representing a structured description of future risks.

[0053] The updated dynamic failure network model is used as the initial state, where each node has an initial real-time risk value and directed edges have current weights. Risk extrapolation algorithms based on complex network theory can be used, such as Monte Carlo simulations or dynamic models based on differential equations. The algorithm simulates how risk propagates step-by-step through the network within a given future time period, starting from currently high-risk nodes and following the directions indicated by the directed edges and the probabilities and time delays defined by the edge weights. Through thousands of simulation iterations, the algorithm calculates the probability of each node, i.e., each failure mode, being triggered at the end of that future time period.

[0054] As one implementation method, the algorithm identifies the most frequently occurring chain of events from the initial risk point to the final severe consequences in multiple simulations—the critical failure propagation path. For each link in the path, the algorithm can also estimate the expected timeframe for that link to occur based on statistical analysis of propagation delays. All these calculation results are integrated and formatted to generate a complete risk prediction information report, serving as a direct basis for subsequent classification and decision-making.

[0055] In the implementation of the above embodiments: the performance grading method enables more accurate and forward-looking risk assessment. By dynamically calculating node risk and edge weights, the model can accurately reflect the actual response of the structure in real time, improving the accuracy and timeliness of risk perception. Using the updated model for extrapolation, it is possible to predict potential future failure chains and their timelines, achieving a leap from current situation perception to future early warning.

[0056] Optionally, in the embodiments of this application, the failure modes represented by the nodes in the dynamic failure network model include at least one of track slab misalignment, support slippage, main beam displacement exceeding limits, pier bending cracking, pier shear failure, and beam collapse.

[0057] Each of the above failure modes requires a clear engineering definition before modeling and must be associated with observable physical quantities or state thresholds. The following is an explanation of each risk mode: Track slab misalignment: refers to the lateral or longitudinal offset of the track relative to the design position. Its occurrence can be defined by measuring the track gauge and track orientation data through a geometric monitoring system deployed on the track.

[0058] Support slippage: refers to the planar displacement of a bridge support that exceeds the design allowable range. Displacement sensors are typically installed at the supports to monitor the relative displacement between the bridge structure and the piers, and the displacement is compared to a slippage threshold.

[0059] Main beam displacement exceeding limits: This refers to the longitudinal or lateral displacement response of the main beam exceeding a preset safety threshold. This can be monitored in real time using GPS or displacement gauges installed at the beam ends.

[0060] Bridge pier bending cracking and bridge pier shear failure: These are two different failure mechanisms for bridge piers. Bending cracking usually begins with cracks appearing in the concrete on the tension side of the pier base. The risk can be determined by monitoring whether the tensile strain of strain gauges attached to this area exceeds the cracking strain of the concrete. Shear failure, on the other hand, manifests as diagonal cracks, and its risk may be related to the shear strain of the pier body, principal tensile stress, or specific vibration modes observed.

[0061] Beam detachment: refers to the main beam falling off the support or pier. The risk is usually caused by the development of pre-existing conditions such as support slippage and excessive displacement of the main beam. The condition for its occurrence can be defined as the beam end displacement exceeding the ultimate restraint capacity of the support or stop.

[0062] When constructing the initial dynamic failure network model, the key failure modes mentioned above were selected as nodes in the network based on the design drawings of the target bridge, finite element analysis results, and historical earthquake damage data. Each node in the model is not only a name label but also associated with its specific state decision logic or function. These decision logics take real-time state data as input, thereby quantitatively calculating the activation level or probability of occurrence of the node at the current moment during model operation.

[0063] In the implementation of the above embodiments: by clearly defining the failure modes represented by the nodes in the dynamic failure network model, including key damage forms such as track slab misalignment and support slippage, the risk identification and analysis objectives of the performance grading method are very clear. The modeling approach based on specific failure modes facilitates the subsequent transformation of abstract risks into specific seismic performance levels, improving the targeted nature of structural safety protection and overall management efficiency.

[0064] Optionally, in this embodiment, the directed edges between nodes represent the association between failure modes. The association between failure modes refers to the causal triggering relationship between the preceding failure mode and the subsequent failure mode. The preceding failure mode refers to the failure mode corresponding to the starting node of the directed edge; the subsequent failure mode refers to the failure mode corresponding to the ending node of the directed edge.

[0065] In this context, the preceding failure mode is defined as the specific failure mode corresponding to the starting node of the directed edge, such as support slippage, while the subsequent failure mode is the failure mode corresponding to the ending node of the directed edge, such as beam collapse. A directed edge from A to B indicates that the occurrence or development of A (the preceding failure mode) will directly increase the risk of B (the subsequent failure mode) occurring or may directly lead to the occurrence of B.

[0066] One way to establish such causal relationships is as follows: First, when constructing the initial network model, through engineering mechanism analysis, finite element numerical simulation, and statistical analysis of historical earthquake damage cases, identify and determine which failure modes have real physical causal chains. For example, based on mechanical principles, support slippage may lead to the beam displacement not being effectively constrained, thus significantly increasing the risk of beam collapse. Therefore, a directed edge can be established from the support slippage node to the beam collapse node.

[0067] Each directed edge needs to be assigned a quantified attribute, namely a weight, to express the probability and time delay of the excitation relationship. This can be accomplished by combining theoretical models with data analysis. For example, by analyzing a large amount of simulation or experimental data, the conditional probability of shear failure occurring at different time points after the pier bending crack occurs can be statistically determined, thereby assigning initial estimates of the excitation conditional probability and propagation time delay to the directed edges connecting these two points. During the method's operation, the weight parameters of these edges are also dynamically adjusted according to the degree of structural nonlinearity and damage velocity reflected by real-time state data, enabling the causal excitation relationship model to accurately reflect the actual structural behavior under current seismic excitation.

[0068] In the implementation of the above embodiments: Directed edges between nodes are defined to represent the causal relationship between preceding failure modes and subsequent failure modes. The dynamic failure network model can more realistically simulate the inherent logic and chain reaction of bridge damage evolution under strong earthquakes. This transforms the model from a collection of isolated risk points into a system capable of describing how risks are transmitted and evolved. Risk analysis based on this model can not only identify which failure modes currently pose a high risk, but also reveal where the risks originate, along which path they are most likely to spread, and what serious consequences they may ultimately cause. This provides crucial forward-looking information for subsequent intelligent decision-making and hierarchical control.

[0069] Optionally, in this embodiment, the probability information includes the excitation condition probability and the propagation delay; the probability information is dynamically updated based on real-time state data.

[0070] The trigger condition probability is a value between 0 and 1, representing the likelihood that a subsequent failure mode will be triggered given that a preceding failure mode has already occurred. The propagation delay is a time parameter representing the expected average or typical time interval between the occurrence of the preceding failure mode and its triggering of the subsequent failure mode. The probability information is not fixed; its settings are dynamically updated based on real-time state data.

[0071] Dynamic updates to probabilistic information are achieved through the following methods: During model initialization, the excitation conditional probability and propagation delay of each directed edge are assigned initial values ​​based on historical earthquake damage statistics, finite element numerical simulations, or expert experience. During runtime, real-time status data is continuously input, and the built-in update algorithm analyzes this data to determine whether the current seismic characteristics and structural behavior have altered the intensity or urgency of specific causal relationships. For example, if real-time status data shows that the structure is experiencing severe velocity-pulse vibrations, the algorithm may, based on preset rules or a lightweight machine learning model, dynamically increase the excitation conditional probability of the edge from pier bending cracking to pier shear failure and shorten its estimated propagation delay.

[0072] In the implementation of the above embodiments: the excitation conditional probability and propagation delay can be dynamically updated based on real-time state data. This dynamic update mechanism enables the dynamic failure network model to reduce the limitations of static analysis and capture in real time the true impact of changes in seismic characteristics and the accumulation of structural damage on the strength of the causal relationship between failures. The failure propagation path and predicted time derived by the dynamic failure network model will be closer to the actual situation, improving the accuracy and timeliness of risk situation prediction.

[0073] Optionally, in this embodiment of the application, before inputting real-time state data and risk prediction information into the pre-trained deep reinforcement learning agent to generate control commands, the method further includes... In the digital twin simulation environment of the bridge, the observation space, action space and reward function of the intelligent agent are constructed. The observation space includes real-time state data and the state of the dynamic failure network model. The action space is used to define the set of control instructions for the distributed actuator cluster. The reward function aims to minimize the global structural risk and / or damage to bridge components.

[0074] The construction and acquisition of pre-trained deep reinforcement learning agents are carried out in a digital twin simulation environment of a bridge. This is a high-fidelity virtual model that is completely consistent with the geometry, materials, and mechanical properties of the actual bridge, used to safely and cost-effectively simulate the structural response under seismic loading. The first step in training is to construct the three core elements of the agent: observation space, action space, and reward function.

[0075] The observation space defines the environmental states that the agent can perceive, and is set as a combination of simulated real-time state data from sensors and states from a dynamic failure network model embedded in the model. The action space defines all operations that the agent can perform; it is concretized as a set of control instructions, where each instruction corresponds to a specific operation of an actuator in one or more distributed actuator clusters. The reward function is designed to minimize global structural risk and / or bridge component damage; any action taken by the agent in the simulation that reduces risk or damage will receive a positive reward, and vice versa.

[0076] The digital twin environment is excited by near-fault seismic waves, and the agent is trained by reinforcement learning algorithm to obtain a pre-trained deep reinforcement learning agent. The agent has a policy neural network built in, which is used to calculate and output control commands based on the current observation state.

[0077] Using a large number of near-fault seismic waves with near-fault ground motion characteristics as input, a digital twin environment is repeatedly stimulated to simulate various possible earthquake scenarios. In this environment, the agent, as a software algorithm, is trained using reinforcement learning algorithms, such as Proximal Policy Optimization (PPO) or Deep Deterministic Policy Gradient (DDPG). The agent, through its internal policy neural network, attempts to output a control command based on the current observed spatial state. Subsequently, the environment calculates a new structural state and damage based on this command, and feeds back a reward value through a reward function. Through iterative trial and error and learning, the parameters of the policy neural network are continuously adjusted and optimized, enabling the agent to learn a policy that maximizes long-term cumulative rewards—that is, to learn which optimal set of control commands should be issued under what complex risk situations. After training, the network parameters are fixed, resulting in a pre-trained deep reinforcement learning agent that can be deployed in a real-world system.

[0078] In the implementation of the above embodiments: by pre-training a deep reinforcement learning agent in a digital twin simulation environment, a core with highly adaptable and intelligent decision-making capabilities is obtained. The agent undergoes training on massive amounts of earthquake scenarios in the simulation, mastering strategies for generating efficient control commands when faced with complex, nonlinear real-time state data and risk prediction information. This effectively reduces the rigidity and difficulty in handling uncertainty inherent in traditional predefined rule control strategies, thereby significantly improving the command efficiency and collaborative control effect of distributed actuator clusters in real earthquakes. This makes the entire seismic graded control process more intelligent and precise, improving the overall response efficiency and reliability from risk perception to proactive intervention.

[0079] Optionally, in this embodiment of the application, the seismic performance of the bridge is classified based on risk prediction information, including: Based on risk prediction information, calculate at least one risk quantification indicator; the risk quantification indicator is used to characterize the severity of the bridge's risk.

[0080] Risk quantification metrics are one or a set of scalar values ​​used to characterize the severity of risks to a bridge. Their function is to condense complex, multidimensional risk prediction information into a single, intuitively comparable measure. Risk quantification metrics include global risk score, maximum probability of fatal nodes, and / or critical path risk value, etc.

[0081] The global risk integral is obtained by multiplying the real-time occurrence probabilities of all nodes in the dynamic failure network by pre-defined weighted coefficients representing the severity of the failure consequences, and then summing them. Maximum probability of a critical node: This is determined by selecting the value with the highest predicted probability among several catastrophic failure modes, such as beam collapse and pier shear failure. Critical path risk value: This is obtained by selecting the failure propagation path with the highest current probability, multiplying the trigger conditional probabilities of each link in that path, and obtaining the overall probability of the chain occurring.

[0082] Read the occurrence probability and critical path data of each failure mode from the risk prediction information, and calculate the risk quantification index value at the current moment according to the predetermined algorithm, such as the weighted sum or taking the maximum value mentioned above.

[0083] By comparing risk quantification indicators with risk thresholds of different levels, the seismic performance level of the bridge is determined; different seismic performance levels correspond to different control priorities.

[0084] The seismic performance level of a bridge is determined by comparing quantified risk indicators with risk thresholds at different levels. Risk thresholds are a pre-defined set of numerical boundaries used to delineate different risk level ranges. For example, thresholds T1 and T2 can be set, where T1 is less than T2. ​​The calculated quantified risk indicators are compared with these thresholds: if the indicator is lower than T1, it is determined to be at the routine monitoring level; if the indicator is between T1 and T2, it is determined to be at the active intervention level; and if the indicator is higher than T2, it is determined to be at the comprehensive defense level.

[0085] Different seismic performance levels correspond to different control priorities. This means that each level not only has a different name, but also a clear difference in the control objectives, resource allocation strategies, and actuator activation schemes behind it. For example, the priority of the conventional monitoring level is data collection and early warning, and it may not activate or only activate a few actuators; the priority of the comprehensive defense level is the highest, which will instruct all available actuators to work together at the highest intensity.

[0086] In the implementation of the above embodiments: by quantifying risk indicators and comparing them with preset risk thresholds, automated classification of bridge risk status is achieved. The relatively complex risk prediction information obtained in the previous steps is transformed into a clear and explicit seismic performance level decision. Different levels correspond to differentiated control priorities, so that the invocation of the distributed actuator cluster is no longer a single mode, but is matched according to the severity of the risk. This allows the system to be in a low-power monitoring state when the risk is low, saving energy and reducing equipment wear; when the risk increases, it can automatically and promptly switch to a higher-level defense state, mobilizing more control resources for precise intervention.

[0087] Optionally, in this embodiment, the distributed actuator cluster includes at least one of a shape memory alloy limiting device, a self-healing concrete unit, and a magnetorheological damper.

[0088] Shape memory alloy (SMA) restraint devices are intelligent constraint devices made using the properties of SMA materials. They can be installed between the main beam and the pier or abutment. Under normal conditions, they are in a flexible state and do not restrict the normal temperature deformation of the structure. When a specific electrical trigger signal is received, the SMA material undergoes a phase change, instantly restoring its preset rigid shape, thereby forming a strong mechanical constraint between the beam and the pier, actively preventing beam collapse due to excessive displacement.

[0089] Self-healing concrete units are intelligent material modules pre-embedded in crack-prone areas of concrete components such as bridge piers and main beams. They contain numerous microcapsules filled with repair adhesive and a distributed micro-heating network. When the system predicts that the concrete stress at a given location will exceed the cracking threshold, a control command triggers the unit's heating network, causing the microcapsule walls in that specific area to rupture. The adhesive flows out and seeps into the microcracks that are about to form or have already formed, and after curing, it fills and repairs the cracks, restoring some of the material's properties.

[0090] Magnetorheological dampers are intelligent vibration reduction devices with adjustable damping force, which can be installed at support locations or between bridge floors. They are filled with a magnetorheological fluid whose viscosity characteristics (i.e., the magnitude of the damping force) can be precisely controlled in real time by an external magnetic field. The control commands issued by the system include the set value of the required output damping force for each damper. After receiving the command, the damper adjusts the magnetic field strength by adjusting the current in its internal coil, thereby adjusting the damping force to the target value within milliseconds, optimally dissipating seismic input energy and suppressing structural vibration.

[0091] In the implementation of the above embodiments: by configuring a distributed actuator cluster including shape memory alloy limiting devices, self-healing concrete units, and magnetorheological dampers, the magnetorheological dampers provide fast and adjustable vibration control; the shape memory alloy devices provide crucial rigid backup constraints; and the self-healing concrete units provide active repair capabilities for material damage. This combination enables the system to flexibly deploy the most suitable actuators or actuator combinations to implement graded control based on risk prediction information and the determined seismic performance level.

[0092] Optionally, in this embodiment of the application, the method further includes: After the distributed actuator cluster executes the corresponding control commands, real-time feedback data from the bridge is collected.

[0093] Real-time feedback data refers to the latest bridge status data immediately collected by the sensor network after the distributed actuator cluster completes its actions according to the received control commands. Essentially, real-time feedback data is still real-time status data, but specifically refers to the structural response measured after intervention measures are implemented, reflecting the control effect. Examples include beam end displacement changes after damper activation and local strain relief after self-healing unit triggering. This can be achieved through continuous operation of the sensor network. The system synchronously reads the readings of each sensor within a very short time window after sending the control command, serving as a basis for evaluating the effectiveness of the control actions and the latest structural state.

[0094] Real-time feedback data is transmitted to the deep reinforcement learning agent, and the agent is optimized using the real-time feedback data.

[0095] Deep reinforcement learning agents are the decision-making modules that previously generated control commands. Optimization refers to the online fine-tuning or experience accumulation of the decision-making strategies within the agent, i.e., the parameters of its policy neural network, using real-time feedback data.

[0096] For example, the system stores the state before executing a command, the issued command, and the feedback state after execution as a complete experience data package in the agent's experience replay buffer. The agent can periodically use this new data from real earthquake responses to perform additional training iterations on its policy network or to update its understanding model of environmental dynamics. This allows the agent to continuously fine-tune its decisions based on the actual control effects, thereby better adapting to the real characteristics of the bridge and the specific seismic vibrations occurring.

[0097] In the implementation of the above embodiments: by collecting real-time feedback data after control is executed and using it to optimize the deep reinforcement learning agent, and continuously fine-tuning and improving its decision-making strategy based on the actual control effect, the agent can increasingly accurately understand the behavioral characteristics of a specific bridge in an earthquake, thereby improving the accuracy and effectiveness of subsequent control commands.

[0098] Optionally, in this embodiment of the application, collecting real-time status data of bridges in railways near fault seismic zones includes: Real-time status data of bridges in railways near fault seismic zones are collected by a sensor network deployed on the bridges; the sensor network includes at least one of displacement sensors, strain sensors and acceleration sensors.

[0099] A sensor network refers to a distributed set of measuring devices deployed at key locations on a bridge, specifically designed to collect real-time status data of bridges in railways near fault-prone seismic zones. A sensor network includes at least one of displacement sensors, strain sensors, and acceleration sensors. Displacement sensors, such as GPS receivers or wire-type displacement gauges, are used to measure the absolute or relative positional changes of parts such as main beams and piers; strain sensors, such as resistance strain gauges or fiber optic grating sensors, are used to measure the internal deformation of key sections of concrete or steel; and acceleration sensors are used to measure the vibration acceleration response at various points on the structure.

[0100] In the implementation of the above embodiments: a sensor network containing various types of sensors is deployed to collect real-time status data, providing comprehensive raw information input for the subsequent decision-making process of this performance grading method. Displacement, strain, and acceleration data, from different dimensions such as geometric deformation, material internal forces, and kinetic inertia, collectively describe the complete mechanical state of the bridge under seismic loading. This comprehensive data foundation makes the decision-making more accurate.

[0101] Please see Figure 2 The diagram shown is a structural schematic of a seismic performance grading device for railway bridge structures in near-fault strong earthquake zones provided in an embodiment of this application; this embodiment of the application provides a seismic performance grading device 200 for railway bridge structures in near-fault strong earthquake zones, comprising: The data acquisition module 210 is used to acquire real-time status data of bridges in railways near fault seismic zones; the real-time status data reflects at least one of the bridge's displacement, strain value, and acceleration. The risk prediction module 220 is used to input real-time status data into the dynamic failure network model to obtain risk prediction information. The dynamic failure network model is a pre-built weighted directed graph model. The nodes in the dynamic failure network model represent the failure modes of the bridge. The directed edges between nodes represent the correlation between failure modes. The weight of the directed edge represents the probability information of triggering the failure mode it points to. The instruction generation module 230 is used to input real-time status data and risk prediction information into a pre-trained deep reinforcement learning agent to generate control instructions; The execution module 240 is used to classify the seismic performance of the bridge based on risk prediction information and send control commands to the distributed actuator cluster deployed on the bridge so that the distributed actuator cluster can receive and execute the corresponding control commands to perform graded control of the bridge structure.

[0102] It should be understood that this device corresponds to the above-described embodiment of the method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones, and is capable of performing the various steps involved in the above-described method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are omitted here to avoid repetition. The device includes at least one software functional module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.

[0103] Please see Figure 3 The diagram shows a structural schematic of an electronic device provided in an embodiment of this application. An electronic device 300 provided in this application includes a processor 310 and a memory 320. The memory 320 stores machine-readable instructions executable by the processor 310. When the machine-readable instructions are executed by the processor 310, the method described above is performed.

[0104] Figure 3 The components shown can be implemented using hardware, software, or a combination thereof. Electronic device 300 may be a physical device, such as a server or PC, or a virtual device, such as a virtual machine or virtualization container. Furthermore, electronic device 300 is not limited to a single device; it can be a combination of multiple devices or a cluster of numerous devices.

[0105] This application also provides a storage medium storing a computer program, which is executed by a processor to perform the above-described method.

[0106] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0107] This application also provides a computer program product, including computer program instructions, which are executed by a processor to perform the method described above.

[0108] It should be understood that the disclosed apparatus and methods can also be implemented in other ways, given the several embodiments provided in this application. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0109] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0110] The above description is only an optional implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of this application should be covered within the protection scope of the embodiments of this application.

Claims

1. A method for classifying the seismic performance of railway bridge structures in near-fault strong earthquake zones, characterized in that, include: Collect real-time status data of bridges in railways near fault seismic zones; The real-time status data reflects at least one of the bridge's displacement, strain value, and acceleration; The real-time status data is input into the dynamic failure network model to obtain risk prediction information; The dynamic failure network model is a pre-constructed weighted directed graph model. The nodes in the dynamic failure network model represent the failure modes of the bridge; the directed edges between the nodes represent the association between the failure modes; and the weight of the directed edge represents the probability information of triggering the failure mode it points to. The real-time status data and the risk prediction information are input into a pre-trained deep reinforcement learning agent to generate control commands. Based on risk prediction information, the seismic performance of the bridge is classified, and the control commands are sent to a distributed actuator cluster deployed on the bridge, so that the distributed actuator cluster receives and executes the corresponding control commands to perform classified control of the bridge structure.

2. The method according to claim 1, characterized in that, The real-time status data is input into the dynamic failure network model to obtain risk prediction information, including: Based on the real-time status data, the real-time risk value of each node and the weight of each directed edge in the dynamic failure network model are calculated, and the dynamic failure network model is updated to generate an updated dynamic failure network model. The updated dynamic failure network model is used to perform risk simulation and generate the risk prediction information; the risk prediction information is used to represent the probability of occurrence of each failure mode, the failure propagation path and the expected occurrence time in the future period.

3. The method according to claim 1, characterized in that, The failure modes represented by the nodes in the dynamic failure network model include at least one of the following: track slab misalignment, support slippage, main beam displacement exceeding limits, pier bending cracking, pier shear failure, and beam collapse.

4. The method according to claim 1, characterized in that, The directed edges between the nodes represent the association between the failure modes. The association between the failure modes refers to the causal triggering relationship between the preceding failure mode and the subsequent failure mode. The preceding failure mode is the failure mode corresponding to the starting node of the directed edge; the subsequent failure mode is the failure mode corresponding to the ending node of the directed edge.

5. The method according to claim 1, characterized in that, The probability information includes the excitation conditional probability and propagation delay, and is dynamically updated based on the real-time state data.

6. The method according to claim 1, characterized in that, Before inputting the real-time state data and the risk prediction information into the pre-trained deep reinforcement learning agent to generate control commands, the method further includes: In the digital twin simulation environment of the bridge, the observation space, action space, and reward function of the agent are constructed; the observation space includes the real-time state data and the state of the dynamic failure network model; the action space is used to define the set of control instructions for the distributed actuator cluster; and the reward function aims to minimize global structural risk and / or damage to bridge components. The digital twin environment is excited using near-fault seismic waves, and the agent is trained through a reinforcement learning algorithm to obtain the pre-trained deep reinforcement learning agent; wherein, the agent has a pre-built policy neural network for calculating and outputting the control commands based on the current observation state.

7. The method according to claim 1, characterized in that, The seismic performance of the bridge is classified based on risk prediction information, including: Based on the risk prediction information, at least one risk quantification indicator is calculated; the risk quantification indicator is used to characterize the severity of the risk to the bridge. The risk quantification index is compared with risk thresholds of different levels to determine the seismic performance level of the bridge; wherein different seismic performance levels correspond to different control priorities.

8. The method according to claim 1, characterized in that, The distributed actuator cluster includes at least one of the following: shape memory alloy limiting device, self-healing concrete unit, and magnetorheological damper.

9. The method according to claim 1, characterized in that, The method further includes: After the distributed actuator cluster executes the corresponding control command, real-time feedback data of the bridge is collected. The real-time feedback data is transmitted to the deep reinforcement learning agent, and the deep reinforcement learning agent is optimized using the real-time feedback data.

10. The method according to claim 1, characterized in that, Real-time status data of bridges in railways near fault-prone seismic zones was collected, including: Real-time status data of the bridges in the near-fault seismic zone are collected by a sensor network deployed on the bridges; the sensor network includes at least one of displacement sensors, strain sensors and acceleration sensors.

Images