A method, equipment, and medium for evaluating the vibration reduction performance of bridges based on digital twins.

By generating multiple candidate twin branches using digital twin technology, configuring branch weights, establishing a probabilistic seismic demand model, identifying spectral blind zones and predicting their migration, the shortcomings in the characterization of multi-state evolution and parameter uncertainty in existing bridge seismic performance assessments are addressed, and the accurate location of high-risk areas and optimization of resampling tasks are achieved.

CN123115853APending Publication Date: 2026-11-17SUZHOU TONGYAO CIVIL ENG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611261930.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing bridge seismic performance assessment methods are insufficient to fully characterize multi-state evolution trends and their parameter uncertainties, and lack continuous tracking of migration in high-uncertainty spectral regions, resulting in lag and repetitiveness in supplementary calculations.

Method used

By using a digital twin-based approach, multiple candidate twin branches are generated, branch weights are configured, a probabilistic seismic demand model is established, spectral blind spots are identified, a blind spot migration map is constructed, the spectral blind spots in the next assessment cycle are predicted, the supplementary sampling task is optimized, and redundant calculations are reduced.

Benefits of technology

It achieves collaborative characterization of bridge multi-state evolution and parameter uncertainty, accurately locates high-risk areas, prospectively assesses the migration trend of spectral blind zones, improves the targeting of supplementary sampling, and reduces redundant dynamic calculations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115853A_ABST
    Figure CN123115853A_ABST
Patent Text Reader

Abstract

This invention discloses a method, device, and medium for evaluating the seismic performance of bridges based on digital twins, relating to the field of seismic performance evaluation technology. The method includes: collecting operational data of the bridge structure, bearings, and damping devices; updating the basic digital twin model and generating a twin state snapshot; generating candidate twin branches and configuring branch weights; establishing a probabilistic seismic demand model and identifying spectral blind zones; associating spectral blind zones, establishing a blind zone migration map, and predicting the spectral blind zones for the next evaluation period; constructing a prospective sampling region, generating sampling tasks, and removing redundant tasks; performing multi-precision dynamic calculations, updating the probabilistic seismic demand model and branch weights, and outputting the seismic performance evaluation results. This invention achieves collaborative characterization of multi-state evolution and parameter uncertainties, prospective evaluation of spectral blind zone migration trends, and targeted configuration of sampling tasks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of seismic performance assessment technology, and in particular to a method, equipment and medium for assessing the seismic performance of bridges based on digital twins. Background Technology

[0002] Bridge seismic performance assessment is typically based on finite element dynamic models, structural health monitoring data, and ground motion analysis. By identifying the state parameters of the bridge structure, bearings, and damping devices, a bridge calculation model is established. Combined with response spectrum analysis, nonlinear time history analysis, or probabilistic seismic demand analysis, indicators such as main beam displacement, pier curvature, bearing slippage, damping device stroke, and energy consumption are calculated to evaluate the bridge's seismic capacity and performance safety reserves under different seismic inputs.

[0003] However, conventional assessment methods often rely on single-state models for calculations, making it difficult to simultaneously characterize the evolution trend of parameters and the various potential bridge states corresponding to their uncertainties. This may result in an incomplete characterization of the direction of change in vibration reduction performance. At the same time, the analysis samples are usually concentrated in a given seismic motion range, lacking continuous tracking of the migration, splitting and merging processes of high uncertainty spectral regions, which can easily lead to lag and repetition in supplementary calculations. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a bridge vibration reduction performance evaluation method based on digital twins to solve the problems of insufficient characterization of bridge multi-state evolution and parameter uncertainty, and the lag and redundancy of sampling tasks under the condition of spectrum blind zone migration in the existing technology.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for evaluating the seismic performance of bridges based on digital twins, comprising: collecting operational data of the bridge structure, bearings, and seismic isolation devices; updating the basic digital twin model based on the operational data and generating a twin state snapshot; generating multiple candidate twin branches based on the state change trends and parameter uncertainties in multiple consecutive twin state snapshots, and configuring branch weights for each candidate twin branch; establishing a probabilistic seismic demand model for each candidate twin branch, and determining the spectral blind zone of each candidate twin branch based on the model cognitive uncertainty, performance failure risk, and response sensitivity output by the probabilistic seismic demand model; associating the spectral blind zones under different state versions, establishing a blind zone migration map, and predicting the spectral blind zone for the next evaluation cycle based on the blind zone migration map; constructing a look-ahead sampling region based on the predicted spectral blind zone, branch weights, and migration uncertainty, generating sampling tasks according to the expected information gain, performance failure risk, and computational cost, and removing redundant sampling tasks based on the similarity of the digital twin response; performing multi-precision dynamic calculations on the sampling tasks, updating the probabilistic seismic demand model and branch weights using the sampling results and newly added operational data, and outputting the bridge seismic performance evaluation results.

[0007] As a preferred embodiment of the bridge vibration reduction performance evaluation method based on digital twins described in this invention, the generation of the twin state snapshot includes: establishing a basic digital twin model containing the main beam, piers, bearings, vibration reduction devices, and foundation boundaries, and establishing the mapping relationship between physical measurement points and model nodes and elements; collecting operational data and environmental data of the bridge structure, bearings, and vibration reduction devices, performing time alignment, anomaly identification, and missing data processing on the operational data, and extracting bridge observation features; updating the equivalent stiffness of the piers, tangential stiffness of the bearings, bearing friction parameters, damping coefficient of the vibration reduction device, velocity index of the vibration reduction device, and equivalent constraint stiffness of the foundation using a constrained weighted least squares method to obtain the state parameter covariance; and encapsulating the state parameters, state parameter covariance, environmental data, operational data time range, frozen parameter identifier, and model version to generate a twin state snapshot with a state version identifier.

[0008] As a preferred embodiment of the bridge vibration reduction performance evaluation method based on digital twins described in this invention, the generation of multiple candidate twin branches includes: reading at least five consecutive twin state snapshots, using environmental regression relationships to eliminate the influence of environmental factors on state parameters, obtaining long-term state residuals, and determining the rate of change and confidence interval of state parameters based on the long-term state residuals; using the current basic digital twin model as the parent model, generating a baseline evolution branch based on the rate of change of state parameters, and generating an accelerated degradation branch based on the endpoints of the confidence interval of the risk direction; retaining a stable maintenance branch when the state parameter changes in three consecutive evaluation cycles do not exceed the parameter noise band, and generating a local anomaly branch when the local monitoring residual reaches the anomaly judgment condition; performing physical rationality verification on the candidate twin branches, and merging the two candidate twin branches when the standardized state parameter distance between two candidate twin branches is less than the state parameter distance threshold and the bridge response similarity is greater than the bridge response similarity threshold; configuring branch weights based on the degree of fitting of operating data, the probability of state trend continuation, and the physical constraint pass rate, and normalizing the branch weights.

[0009] As a preferred embodiment of the bridge seismic performance evaluation method based on digital twins described in this invention, determining the spectral blind zone includes: extracting peak ground acceleration, peak velocity, dominant period of response spectrum, pulse period, effective duration, spectral energy of sensitive period bands, and the ratio of vertical to horizontal ground motion intensity from the ground motion records, and robustly standardizing the ground motion characteristics; generating a set of candidate points shared by each candidate twin branch in the standardized ground motion characteristic space, and establishing a Gaussian process probabilistic seismic demand model for each candidate twin branch and bridge performance requirements; determining model cognitive uncertainty, performance failure risk, and response sensitivity based on the model cognitive uncertainty, performance failure risk, and response sensitivity; weighting the model cognitive uncertainty, performance failure risk, and response sensitivity to obtain a spectral blind zone score, and determining candidate points whose spectral blind zone scores and performance failure risks reach the corresponding thresholds as blind zone candidate points; establishing a nearest neighbor graph of the blind zone candidate points, and determining the connected components of the nearest neighbor graph containing no less than five blind zone candidate points as the spectral blind zone.

[0010] As a preferred embodiment of the bridge seismic performance evaluation method based on digital twins described in this invention, the step of predicting the spectral blind zone for the next evaluation period based on the blind zone migration map includes: mapping the spectral blind zones of adjacent state versions to the same standardized ground motion feature space; calculating the normalized center distance, overlap of candidate point sets, bridge response similarity, and failure mode differences between spectral blind zones; and obtaining a weighted comprehensive matching cost; establishing a main matching relationship based on the comprehensive matching cost and bridge response similarity; determining a splitting relationship based on the association between one spectral blind zone and at least two spectral blind zones; determining a merging relationship based on the association between at least two spectral blind zones and one spectral blind zone; establishing a blind zone migration map with spectral blind zones as nodes and the main matching relationship, splitting relationship, and merging relationship as directed edges; and predicting the spectral blind zone center, prediction uncertainty, bandwidth, area, and risk level for the next evaluation period based on the spectral blind zone center and migration speed in the continuous state versions.

[0011] As a preferred embodiment of the bridge seismic performance evaluation method based on digital twins described in this invention, the construction of the prospective sampling region includes: mapping the predicted spectral blind zones of each candidate twin branch to a unified candidate point set; determining the multi-branch weighted coverage based on branch weights, spectral blind zone coverage status, and performance failure risk; dividing the candidate points of ground motion characteristics into core sampling regions, branch sampling regions, migration front regions, split buffer zones, and boundary correction regions in sequence based on multi-branch weighted coverage, the number of covered branches, spectral blind zone migration direction, splitting relationship, and spectral blind zone boundary; determining candidate sampling points within each prospective sampling region; normalizing the model cognitive uncertainty, performance failure risk, expected information gain, regional bonus, and computational cost; and calculating the sampling priority; and generating sampling tasks corresponding to the candidate twin branches and ground motion inputs according to the sampling priority.

[0012] As a preferred embodiment of the bridge vibration reduction performance evaluation method based on digital twins described in this invention, the step of removing redundant sampling based on digital twin response similarity includes: obtaining dimensionless bridge response vectors corresponding to candidate ground motions using a reduced-order digital twin model; determining bridge response similarity based on the distance between bridge response vectors; retaining candidate ground motions with higher sampling priority when the bridge response similarity is not less than the response repetition judgment threshold; merging corresponding sampling tasks into a shared sampling task when the standardized state parameter distance and bridge response similarity of different candidate twin branches reach the sampling task merging condition; performing preliminary dynamic calculations using a reduced-order digital twin model; upgrading to a medium-precision digital twin model when the performance demand ratio reaches the first performance demand ratio threshold; determining the performance demand ratio reaching the second performance demand ratio threshold, the bearing entering a slip state, and the damping device stroke utilization rate reaching the stroke utilization rate threshold as high-precision upgrade trigger states; upgrading to a high-precision digital twin model after detecting a high-precision upgrade trigger state; and binding the sampling task with the state version, candidate twin branches, ground motion samples, and model version.

[0013] As a preferred embodiment of the bridge seismic performance evaluation method based on digital twins described in this invention, the step of updating the probabilistic seismic demand model and branch weights using supplementary sampling results and newly added operational data includes: writing supplementary sampling results consistent with the current version into the dataset of the corresponding candidate twin branch, updating the probabilistic seismic demand model, and determining the supplementary sampling termination state based on the degree of reduction in model cognitive prediction variance, the stability of the candidate point set in the spectral blind zone, and the change in performance demand prediction; collecting newly added operational data, calculating the residuals between the predicted response of each candidate twin branch and the newly added operational data, determining the branch matching likelihood based on the residuals, and updating and normalizing the branch weights; disabling candidate twin branches whose branch weights are lower than the disabling threshold for multiple consecutive evaluation periods; when the maximum branch weight is lower than the validity judgment threshold, and the predicted response residuals of each candidate twin branch exceed the allowable range, generating a new local anomaly branch based on the bridge local unit with the largest residual; and outputting bridge seismic performance indicators, performance change trends, main control spectral regions, spectral blind zone migration direction, and seismic device maintenance priority based on the effective candidate twin branches and branch weights.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, the memory storing a computer program, wherein the computer program, when executed by the processor, implements any step of the bridge vibration reduction performance evaluation method based on digital twins as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the bridge vibration reduction performance evaluation method based on digital twins as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by generating multiple candidate twin branches based on continuous twin state snapshots and configuring branch weights, a collaborative characterization of the multi-state evolution and parameter uncertainty of bridges is achieved; by comprehensively recognizing uncertainty, performance failure risk, and response sensitivity through the model, spectral blind zones are identified, enabling accurate positioning of high-risk areas for bridge vibration reduction performance; by establishing a spectral blind zone migration map and predicting the spectral blind zone for the next evaluation cycle, a forward-looking assessment of the migration, splitting, and merging trends of spectral blind zones is achieved; by constructing a forward-looking supplementation region and removing redundant supplementation tasks based on the similarity of digital twin responses, the targeting of supplementation is improved and repetitive dynamic calculations are reduced. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a bridge vibration reduction performance evaluation method based on digital twins.

[0019] Figure 2 A flowchart for generating candidate twin branches and configuring branch weights.

[0020] Figure 3 This is a flowchart for the correlation of spectrum blind zones and prediction of the next evaluation cycle.

[0021] Figure 4 A flowchart for the construction of prospective supplementation regions and the generation of supplementation tasks.

[0022] Figure 5 This is a comparison chart showing the predicted utilization rate of the maximum shock absorber stroke under actual working conditions.

[0023] Figure 6 A comparison chart of prediction biases for spectrum blind zone migration. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Reference Figures 1-6 As one embodiment of the present invention, this embodiment provides a method for evaluating the vibration reduction performance of bridges based on digital twins, including the following steps: S1. Collect operational data of the bridge structure, bearings, and damping devices, update the basic digital twin model based on the operational data, and generate a snapshot of the twin state.

[0028] Furthermore, bridge design data, construction data, vibration damping device product data, historical testing data, and bridge sensor layout information are imported into the digital twin modeling module to establish a digital twin model of the bridge foundation.

[0029] The basic digital twin model includes at least the calculation units corresponding to the main beam, piers, bearings, damping devices, and foundation boundaries, as well as the mapping relationship between the bridge entity measurement points and the nodes, units, and local coordinate directions of the calculation model.

[0030] In one alternative implementation, the basic digital twin model is built using OpenSeesPy, the main beam and piers are represented by spatial beam-column elements, the supports are represented by nonlinear connection elements that can describe initial stiffness, friction slip and limiting clearance, the damping device is represented by velocity-dependent damping elements, and the foundation constraints at the bottom of the piers are represented by equivalent spring elements.

[0031] Furthermore, operational data on the bridge structure, bearings, and vibration damping devices are collected according to the monitoring cycle.

[0032] Among them, the bridge structure operation data includes at least one of the following: main beam acceleration, pier top acceleration, pier strain, main beam displacement, and pier top displacement; the bearing operation data includes at least one of the following: bearing relative displacement, bearing slippage state, and acceleration near the bearing; the damping device operation data includes at least one of the following: relative displacement at both ends of the damping device, relative velocity, and force.

[0033] Furthermore, environmental data related to the bridge's operational status are collected, including at least one of ambient temperature, vehicle load level, wind speed, and foundation water level.

[0034] Based on the bridge sensor installation list, the following data mapping relationship is established: Sensor identification — bridge physical location — digital twin model node or unit — measurement direction — data dimensions.

[0035] For operational data with different sampling frequencies, resampling and time alignment are performed according to a unified time base.

[0036] In this embodiment, the median and median absolute deviation within a sliding window are used to identify isolated outlier data.

[0037] When the deviation of any monitoring data from the median is greater than the product of the outlier multiple and the absolute deviation of the median, the monitoring data is marked as outlier data.

[0038] Optionally, the anomaly factor can range from 3 to 5. If sufficient historical data has been accumulated during the normal operation of the bridge, the anomaly factor can also be determined based on the 99th quantile of the residual distribution of the normal data.

[0039] For data with consecutive missing lengths less than the allowable missing window, shape-preserving interpolation is used for compensation; for data with consecutive missing lengths not less than the allowable missing window, forced interpolation is not performed, but the weight of the corresponding observation in the model update process is reduced, and the covariance of the state parameters related to the observation is increased.

[0040] Furthermore, bridge observation features are extracted from the preprocessed operational data.

[0041] Among them, the bridge observation characteristics include at least one of the following: the first few modal frequencies, modal damping ratio, peak displacement of the main beam, peak relative displacement of the supports, peak force of the damping device, and response spectrum value of the selected measuring point.

[0042] This embodiment uses a constrained weighted least squares method to update the state parameters in the basic digital twin model.

[0043] The status parameters to be updated include at least the equivalent stiffness of the pier, the tangential stiffness of the support, the friction parameters of the support, the damping coefficient of the damping device, the velocity index of the damping device, and the equivalent constraint stiffness of the foundation.

[0044] State parameter update objective function , represented as: ; in, For bridge observation feature vectors, Based on the digital twin model in state parameters The calculated features of the output For the observation weight matrix, These are the state parameters corresponding to the previous state version. This is the parameter covariance matrix corresponding to the previous state version. These are the prior constraint coefficients.

[0045] Among them, the prior constraint coefficients are determined by cross-validation or grid search based on historical monitoring samples to minimize the verification prediction error of the digital twin model, while taking into account the fit of the current observation data and the deviation of the state parameters from the previous state version.

[0046] Furthermore, the sensitivity of the bridge observation characteristics to the state parameters is calculated using finite difference or automatic differentiation, and iterative solutions are performed within the upper and lower bounds of the parameters determined by bridge design data, vibration damping device product data, and historical testing data.

[0047] The state parameter update will terminate when any of the following conditions are met: ; or: ; in, Indicates the number of iterations. These are the status parameters to be updated.

[0048] Updated state parameter covariance , represented as: ; in, This is the sensitivity matrix of bridge observation features to state parameters.

[0049] It should be noted that when the condition number of the sensitivity matrix exceeds the preset upper limit, it indicates that some state parameters are unrecognizable under the current operating data conditions. In this case, the state parameters with lower sensitivity are fixed, only the recognizable state parameters are updated, and the frozen parameter identifiers are recorded.

[0050] Finally, the updated bridge state parameter vector is obtained. The bridge state parameter vector includes the equivalent stiffness of the piers, the tangential stiffness of the supports, the friction parameters of the supports, the damping coefficient of the damping device, the velocity index of the damping device, and the equivalent constraint stiffness of the foundation.

[0051] Furthermore, the updated state parameter vector, state parameter covariance, environmental data, runtime data time range, objective function value, frozen parameter identifier, basic digital twin model version, and dynamic solver version are encapsulated to generate a twin state snapshot and configure a unique state version identifier.

[0052] Ultimately, a basic digital twin model and a snapshot of the twin state corresponding to the current physical state of the bridge are obtained.

[0053] S2. Based on the state change trend and parameter uncertainty in multiple consecutive twin state snapshots, generate multiple candidate twin branches and configure branch weights for each candidate twin branch.

[0054] Furthermore, at least five consecutive twin state snapshots are read, and time series of each state parameter and corresponding environmental data are extracted respectively.

[0055] Furthermore, the changes in recoverable parameters caused by environmental factors are separated from the long-term state changes of the bridge structure or vibration damping device.

[0056] In this embodiment, ambient temperature, vehicle load level, and foundation water level are used as explanatory variables. A Huber robust regression with a quadratic temperature term is employed to establish the environmental regression function: ; in, For the first An environmental regression function with state parameters For ambient temperature, For vehicle load level, Based on the water level, to For the first Environmental regression coefficients for each state parameter.

[0057] Among them, the environmental regression coefficient , represented as: ; in, Here is the Huber loss function. For the first The ambient temperature for each monitoring cycle For the first Vehicle load levels for each monitoring period For the first The baseline water level for each monitoring cycle For the first Updated values ​​of state parameters in a twin state snapshot.

[0058] It should be noted that before establishing the environmental regression function, the ambient temperature, vehicle load level and base water level were standardized based on the mean and standard deviation of historical monitoring data to eliminate the influence of different environmental variables on the solution of regression parameters.

[0059] Subtracting the corresponding environmental regression value from the original state parameters yields the long-term state residuals: ; in, For the first The long-term state residuals of the state parameters after removing environmental influences from a twin state snapshot.

[0060] It should be noted that for bridges where basic water levels or vehicle load levels have not been collected, the corresponding regression terms can be deleted, and environmental regression relationships can be established using only the available environmental data.

[0061] Furthermore, the Theil-Sen median slope is used to estimate the rate of change of the long-term state residuals. And the 90% confidence interval of the rate of change was obtained by residual auto-resampling.

[0062] Let the time interval for the next evaluation period be . Using the current basic digital twin model as the parent model, it generates baseline evolution branches, stability maintenance branches, accelerated degradation branches, and local anomaly branches.

[0063] The state parameters of the baseline evolution branch are expressed as follows: ; in, The first branch of the baseline evolution in the next evaluation period Predicted values ​​of each state parameter.

[0064] The state parameters for maintaining a stable branch are expressed as: ; in, To maintain stability in the next evaluation period, the branch is the first Predicted values ​​of each state parameter.

[0065] Among them, the stability maintenance branch is retained only when the changes in state parameters do not exceed the parameter noise band for three consecutive assessment periods after the removal of environmental impacts.

[0066] The state parameters for accelerating degenerate branches are expressed as: ; in, For the accelerated degradation branch in the next evaluation period, the first Predicted values ​​of each state parameter The confidence interval endpoints on the side where the bridge's vibration reduction performance is reduced.

[0067] For example, when the decrease in the equivalent stiffness of the bridge pier or the damping coefficient of the vibration reduction device is in the direction of risk, the minimum endpoint of the confidence interval is taken; when the increase in the bearing friction parameter is in the direction of risk, the maximum endpoint of the confidence interval is taken.

[0068] Furthermore, local anomaly branches are generated based on the local monitoring residuals.

[0069] If the standardized residual corresponding to a certain support or damping device exceeds the set value, such as 3, for two consecutive monitoring windows, and the standardized residual of adjacent unrelated measuring points does not increase synchronously, then the digital twin unit corresponding to the support or damping device will be identified as an abnormal location.

[0070] Based on the location of the anomaly, generate at least one of the following local anomaly branches, A1-A4: A1: The corresponding state parameter is reduced to the lower confidence bound.

[0071] A2: The bearing friction parameters are increased to the upper confidence limit.

[0072] A3: The damping coefficient of the shock absorber decreases according to a fixed abnormality ratio.

[0073] A4: The remaining stroke of the shock absorber decreases according to the monitored changes.

[0074] When multiple abnormal locations occur simultaneously, independent local abnormal branches are generated first; a joint abnormal branch is generated only when none of the independent local abnormal branches can explain the actual bridge operation data.

[0075] Furthermore, physical rationality checks are performed on each candidate twin branch.

[0076] Among them, the physical rationality verification includes the positive definiteness verification of the overall stiffness matrix of the bridge, the verification of the mass conservation of the bridge, the verification of the connection relationship of components, the verification of the allowable stroke of the bearings, the verification of the rated force of the damping device, the verification of the rated stroke of the damping device, the verification of the range of material parameters, and the verification of the energy stability of dynamic calculation.

[0077] Candidate twin branches that fail any of the necessary physical plausibility checks will not proceed to subsequent spectrum blind zone identification.

[0078] Furthermore, candidate twin branches with similar states are merged.

[0079] When the distance between the standardized state parameters of two candidate twin branches is less than the state parameter distance threshold, and the similarity of the bridge response outputs under the same set of representative small amplitude inputs is greater than the bridge response similarity threshold, the two candidate twin branches are merged.

[0080] Among them, the state parameter distance threshold and the bridge response similarity threshold are calibrated based on the spectral blind zone boundary deviation and the bridge performance requirement deviation before and after the historical candidate twin branches are merged. In this embodiment, the state parameter adopts the root mean square normalized distance. When the root mean square normalized distance is less than 0.2 and the bridge response similarity under the representative small amplitude input is greater than 0.98, the two candidate twin branches are considered to be equivalent in terms of state parameters and dynamic response, and they are merged.

[0081] Furthermore, based on the degree of fit of each candidate twin branch to the most recent running data, the probability of state trend continuation, and the pass rate of physical constraints, the initial branch weights are configured:

[0082] in, For the first The initial branch weights of the candidate twin branches, For the first The residuals between the predicted responses of each candidate twin branch and the most recently run data. The observation error covariance matrix, This represents the probability of the state trend continuing. For physical constraint pass coefficients The sign indicates proportionality.

[0083] The initial branch weights of all candidate twin branches are normalized so that the sum of the initial branch weights of all candidate twin branches is 1.

[0084] It should be noted that the number of candidate twin branches can be selected from 3 to 8.

[0085] When the number of candidate twin branches exceeds the set limit, pruning is performed based on the branch weights and differences in bridge responses.

[0086] Ultimately, multiple candidate twin branches are obtained, each with a clearly defined parent state version, state parameters, branch type, and branch weight.

[0087] S3. Establish probabilistic seismic demand models for each candidate twin branch. Based on the model cognitive uncertainty, performance failure risk, and response sensitivity output by the probabilistic seismic demand models, determine the spectral blind zone of each candidate twin branch.

[0088] Seismic motion records were obtained from natural ground motion databases, ground motion sets used in bridge seismic design, and historical dynamic calculation samples.

[0089] Seismic motion feature vectors are extracted from ground motion records from various locations. These feature vectors include peak ground acceleration, peak ground velocity, dominant period of response spectrum, pulse period, effective duration, spectral energy within the bridge sensitive period zone, and intensity ratio of vertical to horizontal ground motions.

[0090] Furthermore, robust standardization of the interquartile ranges was performed on the seismic characteristics of various regions.

[0091] A Latin hypercube sampling method is employed to generate a unified set of candidate points in a standardized seismic motion feature space. All candidate twin branches use the same set of candidate points to enable direct comparison of the spectral blind zones corresponding to different candidate twin branches.

[0092] Furthermore, a Gaussian process probabilistic seismic demand model was established for each candidate twin branch and each bridge performance requirement.

[0093] Bridge performance requirements include at least one of the following: maximum displacement of the main girder, maximum curvature of the pier, maximum slip of the bearing, maximum stroke of the damping device, cumulative energy dissipation of the damping device, and residual displacement of the bridge.

[0094] In this embodiment, the Gaussian process employs a Matérn process with an anisotropic length scale. Kernel function: ; in, The weighted distance between two seismic motion eigenvectors after considering each eigenscale is expressed as: ; in, For input samples and The kernel function values ​​between This is a standardized seismic motion feature vector. For another standardized seismic motion eigenvector, The square of the kernel function amplitude. Earthquake motion characteristic vector The first in Item features, Earthquake motion characteristic vector The first in Item features, For the first The length scale of the earthquake motion characteristics.

[0095] The kernel function magnitude, length scales of each feature, and observation noise are determined by maximizing the logarithmic marginal likelihood of the Gaussian process.

[0096] It should be noted that separate probabilistic seismic demand models are established for different bridge performance requirements. Candidate twin branches can inherit the kernel parameters of the corresponding probabilistic seismic demand model of the parent branch as initial values, but they need to be re-optimized using the data of the candidate twin branches of the current model.

[0097] Furthermore, the model cognitive uncertainty for each candidate point is calculated.

[0098] Specifically, no. Each candidate twin branch at the candidate point The cognitive uncertainty of the normalized model at that location is: ; in, For the first Candidate twin branches in seismic motion feature vectors The normalized model for cognitive uncertainty has a value range of ≤1. This is a number representing the bridge's performance requirements. For the first In the candidate twin branch, the first Bridge performance requirements in seismic motion characteristic vectors The model's cognitive prediction variance at that location It represents the 95th percentile of the model's cognitive prediction variance across all candidate points.

[0099] Furthermore, the risk of performance failure is calculated based on the bridge performance requirement prediction results and performance limits: ; in, For the first Candidate twin branches in seismic motion feature vectors Performance failure risk at the location The average predicted performance requirements for bridges. For the corresponding performance limits, To observe the noise variance, It is the standard normal cumulative distribution function.

[0100] The performance limits are determined based on bridge design documents, pier section analysis results, bearing product information, vibration damping device product information, and bridge operation and maintenance requirements.

[0101] Furthermore, define the maximum performance requirement ratio: ; in, For the first Candidate twin branches in seismic motion feature vectors The maximum performance requirement ratio at that location.

[0102] Calculate the response sensitivity based on the gradient of the maximum performance requirement ratio in the seismic motion characteristic space: ; in, For the first Candidate twin branches in seismic motion feature vectors Response sensitivity at the location.

[0103] Further, calculate the spectrum blind zone score: ; in, For the first Candidate twin branches in seismic motion feature vectors Spectral blind zone score at the location Weights for model cognitive uncertainty, As a weight for the risk of performance failure, Weights for response sensitivity.

[0104] Among them, the model cognitive uncertainty weight, performance failure risk weight, and response sensitivity weight are determined by grid search and cross-validation based on historical samples to minimize the comprehensive evaluation value of the missed detection rate, false detection rate, and performance prediction error of spectrum blind zone identification. The value range is usually 0~1, and the sum of each weight is 1.

[0105] When the candidate points simultaneously satisfy and When this happens, the current candidate point is identified as a blind zone candidate point.

[0106] in, The threshold for spectrum blind zone scoring is the 85th percentile of historical spectrum blind zone scores. The performance failure risk threshold is determined by traversing candidate thresholds and balancing the false negative and false negative rates of spectrum blind zone identification, based on the out-of-limit results of various performance indicators in historical ground motion samples and verification samples. The value can be selected as 0.20.

[0107] Furthermore, establish candidate points for blind spots. Nearest neighbor graph.

[0108] For example, in this embodiment Let the first The distance from each candidate point to its 8th nearest neighbor is Then, the distances from each candidate point to its 8th nearest neighbor are formed into a distance set, and the 75th percentile of the distance set is taken as the connection radius of the nearest neighbor graph. .

[0109] When two candidate blind spots are neighbors and the distance between them is no greater than [missing value], When the two candidate points are connected, a connecting edge is established.

[0110] Calculate the connected components for the nearest neighbor graph. When a connected component includes at least five candidate points for blind zones, the connected component is identified as a spectral blind zone.

[0111] The center of the spectrum dead zone is represented as: ; in, Center of the spectrum blind zone This is the set of candidate points contained in the spectrum blind zone.

[0112] Furthermore, based on the average exceedance probability of the performance requirements of each bridge within the spectrum blind zone, the main control responses and failure modes of the spectrum blind zone are determined.

[0113] Ultimately, the spectrum blind zone center, boundary, candidate point set, risk level, control response, and failure mode corresponding to each candidate twin branch are obtained.

[0114] S4. Associate the spectrum blind spots under different state versions, establish a blind spot migration map, and predict the spectrum blind spots for the next evaluation cycle based on the blind spot migration map.

[0115] Furthermore, the spectral blind zones in adjacent state versions are mapped to the same standardized ground motion feature space.

[0116] For the spectrum blind zone in the previous state version and the spectrum blind zone in the current state version, calculate the center distance of the spectrum blind zone, the degree of overlap of the candidate point set, the similarity of the bridge response, and the failure mode cost, respectively.

[0117] The center distance of the spectrum blind zone is expressed as: ; in, For the first The first spectrum blind zone and the first Normalized center distance between each spectrum blind zone For the first The center vector of each spectral blind zone in the standardized seismic motion characteristic space For the first The center vector of each spectral blind zone in the standardized seismic motion characteristic space The diagonal length of the standardized spatial bounding box of seismic motion characteristics is expressed as: ; in, Let be the dimension of the seismic motion characteristic space. This is the numbering of the characteristic dimension of seismic motion. The first among all candidate points The maximum value of the characteristic ground motion. The first among all candidate points The minimum value of the characteristic ground motion.

[0118] The intersection-union ratio of the candidate point set for the spectrum blind zone is expressed as: ; in, For the first The first spectrum blind zone and the first Intersection over union (IoU) of candidate point sets for each spectrum blind zone For the first The set of candidate points contained in each spectrum blind zone For the first A set of candidate points contained in a spectrum blind zone.

[0119] The bridge response similarity is represented as: ; in, For the first The first spectrum blind zone and the first Bridge response similarity between spectral blind zones and This is the dimensionless frequency blind zone bridge response vector after conforming to the corresponding performance limits.

[0120] Finally, the overall matching cost is obtained, expressed as: ; in, For the first The first spectrum blind zone and the first The overall matching cost between spectrum blind zones The weights for normalized center distances, The weights for the degree of overlap of the candidate point sets. The weights for the similarity of bridge responses, The weights corresponding to the differences in failure modes.

[0121] When the failure modes of the two spectrum dead zones are the same, When the control components are the same but the performance limit types are different, When the control components and failure modes are different, .

[0122] Among them, the weights of normalized center distance, candidate point set overlap, bridge response similarity, and failure mode difference are determined by grid search and cross-validation based on the spectrum blind zone matching samples confirmed in the historical version, so as to minimize the comprehensive evaluation value of spectrum blind zone error matching rate and missed matching rate. The value range is usually 0~1, and the sum of each weight is 1.

[0123] Furthermore, the Hungarian algorithm is employed to minimize the total matching cost of the spectral blind zone in the neighboring state versions.

[0124] when and At that time, a main matching edge is established between the two spectral blind zones. Among them, To determine the comprehensive matching cost threshold, the threshold is determined by traversing candidate thresholds and balancing the false matching rate and the missed matching rate based on the confirmed spectrum blind zone matching samples in the historical version, so as to minimize the comprehensive evaluation value of the two. A value of 0.55 can be selected. The bridge response similarity threshold is determined by traversing candidate thresholds and minimizing the combined false positive and false negative rates of response-similar spectrum blind zones based on the confirmed spectrum blind zone matching samples in the historical state version. A value of 0.70 can be selected.

[0125] Furthermore, it is determined whether the spectrum blind zone has split or merged.

[0126] Specifically, if one spectrum blind zone in the previous state version and two or more spectrum blind zones in the current state version respectively satisfy... and Furthermore, the overlap between the union of the current multiple spectrum blind zone candidate point sets and the previous spectrum blind zone candidate point set is greater than [missing information]. This confirms that the previous spectrum dead zone has split. The blind zone splitting threshold is determined by traversing candidate thresholds and balancing the splitting false positive rate and false negative rate based on the confirmed spectrum blind zone splitting samples in the historical version, so as to minimize the comprehensive evaluation value of the two. A value of 0.65 can be selected. The threshold for the overlap of spectrum blind zone splits is determined by traversing candidate overlap thresholds and balancing the split false positive rate and false negative rate based on the confirmed spectrum blind zone split samples in the historical version, so as to minimize the comprehensive evaluation value of the two. A value of 0.35 can be selected.

[0127] If multiple spectrum blind zones in the previous state version meet the corresponding conditions with the same spectrum blind zone in the current state version, then it is determined that the spectrum blind zones have been merged.

[0128] Current spectrum blind zones that cannot be matched with historical spectrum blind zones are identified as newly emerging spectrum blind zones.

[0129] Furthermore, the spectrum blind spots in different state versions are used as graph nodes, and the main matching, splitting, or merging relationships are used as directed edges to establish a blind spot migration graph.

[0130] For the same spectral blind zone that exists in at least three consecutive state versions, a constant velocity state model is established based on the center location and migration speed of the spectral blind zone.

[0131] Specifically, the state vector is represented as: ; in, For the first The spectrum blind zone migration state vector corresponding to each state version For the first The coordinates of the spectral blind zone center in the first principal ground motion characteristic direction in each state version. For the first The coordinates of the spectral blind zone center in the direction of the second principal ground motion characteristic in each state version. The migration velocity of the center of the spectral dead zone in the direction of the first principal ground motion characteristic is given. This represents the migration velocity of the center of the spectral blind zone in the direction of the second principal ground motion characteristic.

[0132] It should be noted that the first and second principal ground motion feature directions are determined by performing principal component analysis on the seven-dimensional standardized ground motion features. The migration state model predicts the center and velocity of the blind zone in the two-dimensional principal component space, while retaining the center coordinates and boundary ranges of the other five principal component directions. The boundary of the seven-dimensional prediction spectrum blind zone is reconstructed through inverse principal component transformation to uniquely determine whether the candidate point falls into the prediction spectrum blind zone.

[0133] For example, the first principal ground motion characteristic direction can be the dominant period direction of the response spectrum, and the second principal ground motion characteristic direction can be the pulse period direction.

[0134] The state transition model is expressed as: ; in, For the spectrum blind zone migration state vector of the next state version, For the first The process noise vector between one state version and the next state version. The state transition matrix is ​​expressed as: ; in, This is the time interval between the current state version and the next state version, i.e., the time interval of the next cycle.

[0135] The observation model is represented as: ; in, For the first The actual observed vector of the spectral blind zone center obtained in each state version To observe the noise vector, The observation matrix is ​​represented as: ; Process noise covariance matrix , represented as: ; in, This represents the process noise intensity.

[0136] in, The variance is determined based on the variance of the migration acceleration residuals at the center of the historical spectrum blind zone.

[0137] Observation noise covariance matrix , represented as: ; in, The variance of the observation error at the center of the spectral blind zone along the direction of the first principal ground motion characteristic is given. This represents the variance of the observation error at the center of the spectral blind zone along the direction of the second principal ground motion characteristic.

[0138] in, and The candidate points for the spectral blind zone are determined based on the weighted dispersion of the two principal feature directions.

[0139] The initial covariance is determined based on the fitted residuals of the spectral blind zone center from the first three state versions.

[0140] Kalman filtering is used to predict the center location and covariance of the spectrum blind zone in the next evaluation period; exponential smoothing is used to predict the bandwidth, area and risk level of the spectrum blind zone.

[0141] For split spectrum blind zones, migration sub-trajectories are established for each sub-spectrum blind zone; for newly formed spectrum blind zones, a buffer prediction zone is formed by expanding the current spectrum blind zone boundary by 1.5 times the prediction standard deviation, and long-distance extrapolation without historical basis is not performed.

[0142] Finally, the predicted spectral blind zone of each candidate twin branch in the next evaluation cycle is obtained.

[0143] S5. Construct a look-ahead sampling region based on the predicted spectrum blind zone, branch weights, and migration uncertainties. Generate sampling tasks according to the expected information gain, performance failure risk, and computational cost. Remove redundant sampling tasks based on the similarity of the digital twin response.

[0144] Furthermore, the predicted spectral blind zones of each candidate twin branch are mapped to a unified candidate point set.

[0145] For each candidate point, calculate the multi-branch weighted coverage: ; in, Earthquake motion characteristic vector Multi-branch weighted coverage, The number of the candidate twin branch. For the first Branch weights of each candidate twin branch, For the first Candidate twin branches for seismic eigenvectors Coverage indicator function, For the first Candidate twin branches in seismic motion feature vectors The risk of performance failure at the location.

[0146] Among them, when the candidate point is located at the th When the predicted spectrum blind zone of each candidate twin branch is within the region, ;otherwise, .

[0147] Furthermore, based on multi-branch weighted coverage, branch weights, performance failure risk, and spectrum blind zone migration trajectory, the look-ahead sampling region is divided into a core sampling region, a branch sampling region, a migration leading edge region, a split buffer zone, and a boundary correction region.

[0148] when When the coverage of a candidate point is not less than the multi-branch weighted coverage threshold and is covered by at least two candidate twin branches, the region where the candidate point is located is determined as the core supplementary sampling area. When a candidate point is covered by only one candidate twin branch, and Not less than the candidate twin branch weight threshold and When the value is not less than the performance failure risk threshold, the area where the candidate point is located is determined as the branching and supplementary sampling area.

[0149] Among them, the multi-branch weighted coverage threshold is determined by traversing candidate thresholds and balancing the missed detection rate, false detection rate and computational cost of the core supplementation area based on the actual coverage results of the core high-risk area in the historical supplementation samples, and the selected value is 0.60; the candidate twin branch weight threshold is determined by traversing candidate thresholds and balancing the effective branch omission rate and the low confidence branch retention rate based on the posterior weight of the historical candidate twin branches and their prediction accuracy for the subsequent physical bridge status, and the selected value is 0.20; the performance failure risk threshold is determined by traversing candidate thresholds and balancing the missed detection rate and false detection rate of the high-risk area based on the performance exceeding the limit results in the historical ground motion samples and the verification samples, and the selected value is 0.40.

[0150] The migration leading edge region is located outside the predicted spectrum blind region along the migration velocity direction, and its outward extension distance is expressed as: ; in, This represents the outward expansion distance of the leading edge region of the spectrum dead zone migration. The standard deviation of the prediction for the center location of the spectrum dead zone. is the connection radius of the nearest neighbor graph of candidate points in the spectrum blind zone.

[0151] The split buffer is located between the centers of the predicted split sub-trajectories and covers the predicted confidence ellipse of each sub-trajectory.

[0152] The boundary correction region is located both inside and outside the boundary of the predicted spectrum blind zone. A strip-shaped area of ​​width.

[0153] When the aforementioned regions overlap, the region labels are determined in the order of core replenishment region, split buffer zone, boundary correction region, migration front region, and branch replenishment region.

[0154] Furthermore, target ground motion characteristics are generated within the look-ahead sampling area.

[0155] Prioritize selecting ground motion records from the natural ground motion database that match the target ground motion characteristics. When there are insufficient natural ground motion records, use spectral matching to generate ground motion input, and save the seed ground motion record, scaling factor, target response spectrum, spectral matching parameters, and ground motion generation program version.

[0156] Furthermore, the expected information gain of the candidate supplementary sampling points is calculated: ; in, For the first Candidate twin branches in seismic motion feature vectors The expected information gain at the location, For the first Candidate twin branches in seismic motion feature vectors The model's cognitive prediction variance at that location For the first The observation noise variance of the candidate twin branches corresponding to the probabilistic earthquake demand model.

[0157] Candidate supplementary sampling points at the 1st The complementation priority under each candidate twin branch is expressed as follows: ; in, Earthquake motion characteristic vector In the The priority of supplementation under each candidate twin branch This is a migration leading edge or splitting buffer bonus value. Earthquake motion characteristic vector The estimated computational cost of the resampling task, Weights for model cognitive uncertainty, As a weight for the risk of performance failure, The weights are the expected information gain. Weights added to the migration front or split buffer.

[0158] Among them, the model cognitive uncertainty weight, performance failure risk weight, expected information gain weight, and migration front or split buffer bonus weight are determined by grid search and cross-validation based on historical supplementation tasks, so as to minimize the comprehensive evaluation value of the high-risk area missed detection rate, the prediction error after supplementation, and the unit computation cost. The value range is usually 0~1, and the sum of each weight is 1.

[0159] It should be noted that before calculating the supplementation priority, the model cognitive uncertainty, performance failure risk, expected information gain, and migration front or split buffer addition are normalized to a unified numerical range, and the expected computational cost is converted into a dimensionless relative cost relative to the median computational cost of historical tasks, so as to eliminate the impact of differences in the dimensions and numerical scales of different indicators on the ranking of supplementation tasks.

[0160] For the same candidate point that is covered by multiple candidate twin branches, calculate the unified imputation priority: ; in, Earthquake motion characteristic vector Unified supplementation priority across candidate twin branches.

[0161] Furthermore, it is determined whether different candidate twin branches can share the same supplementary sampling task.

[0162] When the distance between the standardized state parameters of different candidate twin branches is less than the standardized state parameter distance threshold, and the bridge response similarity output by the reduced-order digital twin model is not less than the bridge response similarity threshold, the corresponding supplementary sampling tasks are merged into a shared task; otherwise, supplementary sampling tasks are created for each candidate twin branch.

[0163] It should be noted that the shared sampling task only shares the ground motion preprocessing, load time history structure and common calculation results. Within the same task, the state parameters of each candidate twin branch are used to calculate the branch response sub-results and associate them with the corresponding candidate twin branch identifiers, thereby avoiding cross-branch misuse of calculation results and reducing redundant calculations.

[0164] The standardized state parameter distance threshold is determined based on historical candidate twin branch samples by traversing candidate distance thresholds and taking into account the deviation of state parameters before and after branch merging, the deviation of spectrum blind zone boundary, and the critical bridge response error not exceeding the allowable range. A value of 0.2 can be selected. The bridge response similarity threshold is determined based on historical shared supplementary sample samples by traversing candidate similarity thresholds and balancing the missed judgment rate, erroneous sharing rate, and bridge response error after merging in the shared task. A value of 0.98 can be selected.

[0165] Furthermore, a reduced-order digital twin model is used to calculate the bridge response vector corresponding to the candidate seismic motions: ; in, Let be the bridge response vector. , , , , These are the limits for main beam displacement performance, pier curvature performance, bearing slip performance, damping device stroke performance, and damping device reference energy dissipation value. , , , , These are the maximum displacement of the main beam, the maximum curvature of the key section of the pier, the maximum slip of the support, the maximum stroke of the damping device, and the cumulative energy consumption of the damping device under a single seismic input.

[0166] The bridge response distance for two candidate ground motions is expressed as: ; in, Candidate points for seismic motion eigenvectors With candidate points of seismic motion eigenvectors The corresponding bridge response distance, This is the number of the bridge response vector. For the first The weights of the bridge response indicators Candidate points for seismic motion eigenvectors The first in the bridge response vector Item response value, Candidate points for seismic motion eigenvectors The first in the bridge response vector Item response value.

[0167] It should be noted that the weights of the bridge response indicators were determined using grid search and cross-validation, based on the sensitivity of each response indicator to bridge performance failure determination in historical seismic motion samples. The goal was to achieve the highest accuracy in distinguishing key responses while minimizing the error in calculating response distance. The weights ranged from 0 to 1, and the sum of the weights of all bridge response indicators was equal to 1. Bridge response similarity is represented as: ; in, This represents the similarity of bridge responses corresponding to two seismic motion feature vectors. The square of the bridge response distance between two candidate points of the seismic motion eigenvector is given.

[0168] when If the response is not less than the condition for repeatability, such as 0.95, the bridge response generated by the two candidate ground motions is considered to be repeatable, and only the candidate ground motion with the higher priority of unified sampling is retained.

[0169] Furthermore, based on the ratio of the sampled area to the bridge performance requirements, a reduced-order digital twin model, a medium-precision digital twin model, or a high-precision digital twin model is selected to perform dynamic calculations.

[0170] When any performance requirement ratio predicted by the reduced-order digital twin model is not less than the first performance requirement ratio threshold, the task is upgraded to a medium-precision digital twin model; when any performance requirement ratio predicted by the medium-precision digital twin model is not less than the second performance requirement ratio threshold, the support enters a slip state, or the damping device stroke utilization rate is not less than the damping device stroke utilization rate threshold, the task is upgraded to a high-precision digital twin model.

[0171] Among them, the performance demand ratio threshold is determined by traversing candidate thresholds based on the response errors of the reduced-order model, medium-precision model and high-precision model on historical samples, and taking into account the condition that failure states near the performance limit are not missed and the number of model upgrades is minimized. The selectable value of the first performance demand ratio threshold is 0.70, and the selectable value of the second performance demand ratio threshold is 0.85. The shock absorber stroke utilization rate threshold is determined by traversing candidate thresholds based on the allowable stroke of the shock absorber product, historical dynamic calculation results and high-precision model verification results, and balancing the stroke over-limit missed detection rate and high-precision calculation volume. The selectable value is 0.80.

[0172] It should be noted that the reduced-order digital twin model is established by preserving the dominant vibration mode and establishing equivalent parameterization of the supports and damping devices. The medium-precision digital twin model preserves the nonlinear behavior of key components, and the high-precision digital twin model preserves complete elements, degrees of freedom, and nonlinear connection relationships. The three types of models use the same state parameters, boundary conditions, and seismic motion inputs. The calculation results of the high-precision digital twin model are used as the benchmark, and the response correction coefficients of the reduced-order and medium-precision digital twin models are calibrated by representative seismic motion samples. When the model is upgraded, the state version, candidate twin branches, and seismic motion inputs of the original sampling task are inherited and recalculated. The upgraded and corrected calculation results are used to overwrite the original results for subsequent response similarity judgment, redundant task processing, and bridge damping performance output.

[0173] High-precision digital twin models are preferred for replenishment tasks in the core replenishment area and the split buffer.

[0174] Each sampling task is bound to a status version identifier, a candidate twin branch identifier, a seismic motion sample identifier, a bridge parameter file hash value, a digital twin model level, a dynamic solver version, and a task creation time.

[0175] When the resample task is completed, if the state parameters of the parent state version have changed, the calculation results will be saved to the original state version archive area and will not be written to the current probabilistic earthquake demand model training set.

[0176] Finally, the supplementary calculation results corresponding to each candidate twin branch and state version are obtained.

[0177] S6. Perform multi-precision dynamic calculations to complete the resampling task, update the probabilistic earthquake demand model and branch weights using the resampling results and newly added running data, and output the bridge seismic performance evaluation results.

[0178] Furthermore, the results of the supplementary sampling calculation are written into the dataset of the corresponding candidate twin branch, and the probabilistic earthquake demand model of the candidate twin branch is retrained.

[0179] Recalculate the model cognitive prediction variance and spectral blind zone boundary within the prospective supplementation area.

[0180] The supplementation of samples in the current lookahead region will be terminated when all of the following conditions B1-B3 are met: B1: The average reduction rate of the variance in local model cognitive predictions is less than 5%.

[0181] B2: The intersection-union ratio of the candidate point sets of two consecutive rounds of spectrum blind zones is greater than 0.95.

[0182] B3: The change in the mean of the control performance requirement forecast is less than 3%.

[0183] It should be noted that when the computational budget reaches its limit, only the remaining sample replenishment tasks are paused, and the probabilistic seismic demand model is not considered to have converged.

[0184] Furthermore, the operating data of the bridge structure, bearings, and damping devices for the next monitoring cycle are obtained, and the predicted observation vectors corresponding to the current bridge operating conditions are calculated using each candidate twin branch.

[0185] The candidate twin branch matching likelihood is calculated based on the residual between the predicted observation vector and the actual bridge operation data.

[0186] Match the likelihood of the candidate twin branches and update the branch weights of each candidate twin branch.

[0187] When the branch weight of a candidate twin branch is lower than the candidate twin branch deactivation threshold for three consecutive evaluation periods, the candidate twin branch is deactivated.

[0188] When the weight of the largest branch among all candidate twin branches is still lower than the threshold for determining the validity of the candidate twin branch, and the predicted response residuals of all candidate twin branches exceed the allowable range, return to step S2 and generate a new local anomaly branch based on the bridge local unit with the largest standardized residual.

[0189] The candidate twin branch disabling threshold is determined by traversing the candidate thresholds and balancing the invalid branch retention rate and the effective branch erroneous disabling rate, based on the posterior weights of historical candidate twin branches and their subsequent matching results with the actual bridge status. A value of 0.05 can be selected. The candidate twin branch validity judgment threshold is determined by traversing the candidate thresholds and balancing the abnormal status omission rate and the candidate twin branch regeneration frequency, based on the maximum branch weight, predicted response residual, and actual status identification results in the historical evaluation cycle. A value of 0.40 can be selected.

[0190] Furthermore, candidate twin branches whose branch weights meet the retention criteria are identified as valid twin branches.

[0191] Based on the effective twin branch calculation, the main beam displacement damping rate, pier curvature requirement reduction rate, bearing slippage exceeding probability, damping device maximum stroke utilization rate, damping device cumulative energy consumption, bridge residual displacement, performance limit exceeding probability, and damping performance retention rate in the next evaluation cycle are calculated.

[0192] In one alternative implementation, a reference digital twin model is established that has the same bridge structural state, support state, and foundation boundary as the effective twin branch.

[0193] The original connection positions at both ends of the target damping device are kept unchanged by referring to the digital twin model. The energy dissipation parameters of the target damping device unit are set to zero, and the axial initial stiffness corresponding to the physical connection structure of the target damping device is retained.

[0194] It should be noted that the target damping device will not be replaced with a rigid connection to avoid altering the original boundary conditions of the bridge.

[0195] The same seismic input is applied to the effective twin branch and the reference digital twin model, and the net damping contribution of the target damping device is determined based on the difference in bridge response between the two.

[0196] Furthermore, the vibration reduction performance indicators of multiple effective twin branches are weighted according to the branch weights to generate the bridge's current vibration reduction performance level, performance change trend in the next evaluation cycle, main control spectrum region, spectrum blind zone migration direction, and priority for inspection, maintenance, or replacement of vibration reduction devices.

[0197] Thus, a bridge vibration reduction performance assessment based on digital twins was completed.

[0198] In this embodiment, Figure 5 The comparison between the predicted and measured results of the maximum stroke utilization rate of the vibration damping device under field conditions is shown. Specifically, the prediction points of the method of this invention are generally closely distributed on both sides of the ideal prediction line, maintaining a small deviation under different working conditions of low stroke utilization rate, medium stroke utilization rate, and near-high stroke utilization rate; the prediction points of the conventional single-model method have a larger degree of dispersion and show significant overestimation or underestimation under some working conditions.

[0199] The results show that this invention continuously collects operational data from the bridge structure, bearings, and damping devices to update the basic digital twin model. It also uses multiple candidate twin branches to characterize the bridge's state change trends and parameter uncertainties, thus reducing the impact of incorrect assumptions about the state of a single model on the prediction results. Furthermore, by employing supplementary sampling tasks and multi-precision dynamic calculations to specifically correct high-risk areas, the predicted maximum stroke utilization rate of the damping devices becomes closer to the actual operating state, thereby improving the accuracy of identifying excessive stroke, judging performance degradation, and determining the priority of inspection and maintenance for the damping devices.

[0200] In this embodiment, Figure 6 The results show the changes in prediction deviations of the method of this invention and the conventional static evaluation method for the center location of the spectral blind zone over twelve consecutive evaluation periods. Specifically, the prediction deviation of the method of this invention gradually decreased from approximately 11.8% in the first evaluation period to approximately 5.2% in the twelfth evaluation period, showing a generally stable downward trend. While the prediction deviation of the conventional static evaluation method also decreased, it remained at approximately 10.3% in the twelfth evaluation period, significantly higher than that of the method of this invention.

[0201] The results show that this invention establishes a blind zone migration map containing migration, splitting, and merging relationships by associating the spectral blind zones under different state versions. Furthermore, by combining the center location of the spectral blind zone, migration velocity, and prediction covariance, it can make forward predictions for the next evaluation cycle, effectively tracking the dynamic migration of the spectral blind zone as the bridge state changes. Simultaneously, by continuously supplementing key samples through forward-looking sampling regions, expected information gain, and multi-precision dynamic calculations, it reduces model cognitive uncertainty and avoids the reliance of conventional static evaluations on historical fixed samples, thereby improving the reliability of identifying the main control spectral regions and predicting the trend of vibration reduction performance in the next cycle.

[0202] This embodiment also provides a computer device applicable to the bridge vibration reduction performance evaluation method based on digital twins, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the bridge vibration reduction performance evaluation method based on digital twins as proposed in the above embodiment.

[0203] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0204] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the bridge vibration reduction performance evaluation method based on digital twins as proposed in the above embodiments. 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 Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0205] In summary, this invention achieves a collaborative characterization of the multi-state evolution and parameter uncertainty of bridges by generating multiple candidate twin branches based on continuous twin state snapshots and configuring branch weights; it accurately locates high-risk areas of bridge vibration reduction performance by comprehensively recognizing uncertainty, performance failure risk, and response sensitivity through the model; it provides a forward-looking assessment of the migration, splitting, and merging trends of the spectrum blind zone by establishing a spectrum blind zone migration map and predicting the spectrum blind zone in the next evaluation cycle; and it improves the targeting of the supplementary sampling and reduces repetitive dynamic calculations by constructing a forward-looking supplementary sampling region and removing redundant supplementary sampling tasks based on the similarity of the digital twin response.

[0206] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for evaluating the vibration reduction performance of bridges based on digital twins, characterized in that, include: Collect operational data on the bridge structure, bearings, and damping devices; update the basic digital twin model based on the operational data; and generate a snapshot of the twin's state. Based on the state change trends and parameter uncertainties in multiple consecutive twin state snapshots, multiple candidate twin branches are generated, and branch weights are configured for each candidate twin branch; A probabilistic seismic demand model is established for each candidate twin branch. Based on the model cognitive uncertainty, performance failure risk and response sensitivity output by the probabilistic seismic demand model, the spectral blind zone of each candidate twin branch is determined. Associate the spectrum blind spots under different state versions, establish a blind spot migration map, and predict the spectrum blind spots in the next evaluation cycle based on the blind spot migration map; The prospective sampling region is constructed based on the predicted spectrum blind zone, branch weights and migration uncertainties. Sampling tasks are generated according to the expected information gain, performance failure risk and computational cost. Redundant sampling tasks are removed based on the similarity of digital twin responses. The system performs multi-precision dynamic calculations to perform resample tasks, updates the probabilistic seismic demand model and branch weights using the resample results and newly added operational data, and outputs bridge seismic performance assessment results.

2. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 1, characterized in that, The generation of the twin state snapshot includes: Establish a digital twin model of the foundation, including the main beam, piers, bearings, damping devices, and foundation boundaries, and establish the mapping relationship between the physical measurement points and the model nodes and elements; Collect operational data and environmental data of bridge structure, bearings, and damping devices; perform time alignment, anomaly identification, and missing data processing on the operational data; and extract bridge observation features. The constrained weighted least squares method is used to update the equivalent stiffness of the pier, the tangential stiffness of the support, the friction parameters of the support, the damping coefficient of the damping device, the velocity index of the damping device, and the equivalent constraint stiffness of the foundation to obtain the covariance of the state parameters. The state parameters, state parameter covariance, environmental data, runtime data time range, frozen parameter identifier, and model version are encapsulated to generate a twin state snapshot with a state version identifier.

3. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 2, characterized in that, The generation of multiple candidate twin branches includes: Read at least five consecutive twin state snapshots, use environmental regression to remove the influence of environmental factors on state parameters, obtain long-term state residuals, and determine the rate of change and confidence interval of state parameters based on long-term state residuals; Using the current basic digital twin model as the parent model, a baseline evolution branch is generated based on the rate of change of state parameters, and an accelerated degradation branch is generated based on the endpoints of the confidence interval of the risk direction. When the state parameter changes do not exceed the parameter noise band for three consecutive evaluation cycles, the stable maintenance branch is retained; when the local monitoring residual reaches the anomaly judgment condition, the local anomaly branch is generated. Physical rationality verification is performed on the candidate twin branches. When the distance between the standardized state parameters of two candidate twin branches is less than the state parameter distance threshold and the bridge response similarity is greater than the bridge response similarity threshold, the two candidate twin branches are merged. Configure branch weights based on the degree of fit of the running data, the probability of the state trend continuing, and the physical constraint pass rate, and normalize the branch weights.

4. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 1, characterized in that, Determining the spectrum dead zone includes: Peak acceleration, peak velocity, dominant period of response spectrum, pulse period, effective duration, spectral energy of sensitive period band, and intensity ratio of vertical to horizontal ground motion are extracted from the ground motion record, and the ground motion characteristics are robustly standardized. A set of candidate points shared by all candidate twin branches is generated in the standardized seismic motion feature space, and a Gaussian process probabilistic seismic demand model is established for each candidate twin branch and the performance requirements of the bridge. Based on the model cognitive prediction variance, performance demand prediction results, and maximum performance demand ratio gradient, the model cognitive uncertainty, performance failure risk, and response sensitivity are determined respectively. The model cognitive uncertainty, performance failure risk and response sensitivity are weighted to obtain the spectrum blind zone score. Candidate points whose spectrum blind zone score and performance failure risk reach the corresponding threshold are identified as blind zone candidate points. Construct a nearest neighbor graph of candidate blind zone points, and determine the connected components of the nearest neighbor graph containing no less than five candidate blind zone points as spectral blind zones.

5. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 1 or 4, characterized in that, The method of predicting the spectrum blind zone for the next evaluation cycle based on the blind zone migration map includes: The spectral blind zones of adjacent state versions are mapped to the same standardized ground motion feature space. The normalized center distance, the degree of overlap of candidate point sets, the similarity of bridge response and the difference in failure modes between the spectral blind zones are calculated, and the comprehensive matching cost is obtained by weighting. The main matching relationship is established based on the comprehensive matching cost and the bridge response similarity; the splitting relationship is determined based on the correlation between one spectrum blind zone and no less than two spectrum blind zones; and the merging relationship is determined based on the correlation between no less than two spectrum blind zones and one spectrum blind zone. Using spectrum blind zones as nodes and main matching relationships, splitting relationships, and merging relationships as directed edges, a blind zone migration graph is constructed. Based on the spectrum blind zone center and migration speed in the continuous state version, predict the spectrum blind zone center, prediction uncertainty, bandwidth, area and risk level for the next assessment period.

6. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 1, characterized in that, The construction of the prospective supplementary sampling region includes: The predicted spectrum blind zone of each candidate twin branch is mapped to a unified candidate point set, and the multi-branch weighted coverage is determined based on the branch weight, spectrum blind zone coverage status and performance failure risk. Based on multi-branch weighted coverage, number of coverage branches, direction of spectral blind zone migration, splitting relationship, and boundary of spectral blind zone, candidate points of ground motion characteristics are sequentially divided into core sampling area, branch sampling area, migration front area, splitting buffer zone, and boundary correction area. Candidate replenishment points are determined in each prospective replenishment region. The model cognitive uncertainty, performance failure risk, expected information gain, regional bonus and computational cost are normalized, and the replenishment priority is calculated. Generate supplementary sampling tasks corresponding to candidate twin branches and seismic inputs according to the supplementary sampling priority.

7. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 1, characterized in that, The task of removing redundant samples based on the similarity of digital twin responses includes: A reduced-order digital twin model is used to obtain the dimensionless bridge response vectors corresponding to candidate ground motions, and the bridge response similarity is determined based on the distance between the bridge response vectors. When the bridge response similarity is not less than the response repetition judgment threshold, candidate ground motions with higher sampling priority are retained; when the standardized state parameter distance and bridge response similarity of different candidate twin branches reach the sampling task merging condition, the corresponding sampling tasks are merged into a shared sampling task. A preliminary dynamic calculation is performed using a reduced-order digital twin model. When the performance requirement ratio reaches the first performance requirement ratio threshold, it is upgraded to a medium-precision digital twin model. The performance requirement ratio reaches the second performance requirement ratio threshold, the support enters the sliding state, and the stroke utilization rate of the damping device reaches the stroke utilization rate threshold, respectively, as the high-precision upgrade trigger state. After the high-precision upgrade trigger state is detected, it is upgraded to the high-precision digital twin model. The supplementary sampling task is bound to the state version, candidate twin branches, ground motion samples, and model version.

8. The bridge vibration reduction performance evaluation method based on digital twin as described in claim 7, characterized in that, The process of updating the probabilistic earthquake demand model and branch weights using the supplementary sampling results and newly added operational data includes: Write the resampled results consistent with the current version into the dataset of the corresponding candidate twin branch, update the probabilistic earthquake demand model, and determine the resampled termination state based on the degree of reduction in model cognitive prediction variance, the stability of the candidate point set in the spectral blind zone, and the change in performance demand prediction. Collect newly added running data, calculate the residual between the predicted response of each candidate twin branch and the newly added running data, determine the branch matching likelihood based on the residual, and update and normalize the branch weights. Candidate twin branches whose branch weights are below the deactivation threshold for multiple consecutive evaluation periods are deactivated; when the weight of the largest branch is below the validity determination threshold and the predicted response residuals of each candidate twin branch exceed the allowable range, a new local anomaly branch is generated based on the bridge local unit with the largest residual. Based on the valid candidate twin branches and branch weights, the output bridge vibration reduction performance indicators, performance change trends, main control spectrum regions, spectrum blind zone migration direction, and vibration reduction device maintenance priority are provided.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the bridge vibration reduction performance evaluation method based on digital twins as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the bridge vibration reduction performance evaluation method based on digital twins as described in any one of claims 1 to 8.