Cluster learning Kriging-driven bridge digital twin modeling updating system and method

Abstract

The invention belongs to the technical field of digital twinning, and particularly provides a cluster learning Kriging-driven bridge digital twinning modeling updating system and method, and the system comprises a data collection and processing module which is used for obtaining the performance feature information of an actual bridge structure; the digital twinborn model construction module is used for constructing a bridge digital twinborn model and processing to obtain performance characteristic information of a simulated bridge structure; the objective function construction module is used for constructing a residual objective function and obtaining an optimization variable and a parameter space of the optimization variable based on the residual objective function; the optimization module is used for guiding the Kriging agent model to search the optimal parameter in the parameter space by utilizing a cluster optimization mechanism driven by multiple learning functions in parallel; and the model updating module is used for updating the bridge digital twin model through the optimal parameters. According to the method, the bridge structure state recognition precision is effectively improved, the model updating cost is reduced, and the method is suitable for bridge performance prediction and operation and maintenance management under complex working conditions.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of digital twinning, and particularly relates to a cluster learning Kriging driven bridge digital twinning modeling updating system and method. BACKGROUND

[0002] As a long-period service infrastructure, the bridge gradually degrades in structural performance under the continuous action of operating load and environmental change, and the uncertainty of service state increases, which brings challenges to safety assessment and operation management, making it more important to accurately simulate and evaluate the actual state of the bridge. Although high-fidelity numerical simulation has been widely used in bridge structure analysis, it is usually based on idealized design parameters and is difficult to accurately represent the true state of the bridge during long-term service.

[0003] To solve the above problems, digital twinning technology can realize accurate dynamic perception and evolution simulation of the structure by constructing a virtual mapping model of the physical system and relying on twinning updating. However, due to the size and complexity of the bridge structure, the twinning updating process often involves large-scale numerical simulation and consumes a large amount of computing resources. To improve the updating efficiency, existing methods often use surrogate models to replace the original simulation to reduce the computational cost. However, this strategy is significantly dependent on the modeling accuracy of the surrogate model, and may not converge to the true state, thereby affecting the quality of the twinning modeling. Some scholars have distributed optimization tasks to multiple parallel computing nodes to directly solve the original model based on cluster computing resources, which has saved computing resources to some extent while ensuring the accuracy of the results, but the response speed is still difficult to meet the needs of high-frequency data fusion, and the updating efficiency still needs to be further improved.

[0004] Kriging, as a classic Bayesian surrogate model, has been extended as an iterative optimization tool driven by active learning functions in recent years. This mechanism improves the identification and approximation ability of the model in key areas by guiding sampling, and has achieved good application results in some optimization tasks. However, existing methods are mostly based on a single learning function and perform sampling in series, generating only one sample point for surrogate model learning and updating in each iteration, and failing to take advantage of parallel learning and multi-point collaborative optimization potential. Some studies have tried to introduce a multi-objective optimization mechanism to generate multiple candidate sample points in each iteration to support parallel sampling, but the essence is still based on the objective function constructed under the same learning strategy, and has not broken through the single-strategy search paradigm. The sampling distribution is still biased, making it difficult to cover the complex design space comprehensively, limiting its application potential in high-dimensional non-convex and multi-extreme value structure optimization problems.

[0005] In the high-dimensional parameter space of bridge digital twin modeling update, the existing optimization strategy is still difficult to balance the modeling accuracy and update response efficiency in the practical application scenario with complex response characteristics. Therefore, it is urgent to build a cluster optimization method with multi-dimensional parallel collaborative mechanism, break through the sampling guidance limitation under single strategy, and cooperatively drive sample point expansion from multiple optimization perspectives, to improve sampling diversity and model convergence efficiency while ensuring global exploration ability, and provide efficient and stable modeling update capability for bridge digital twin system. SUMMARY

[0006] To solve the problems of existing modeling process such as update dependence on high-fidelity simulation, low sampling efficiency, and limited modeling accuracy, the present application introduces a cluster learning optimization mechanism, combines Kriging surrogate model construction and active learning sample guidance strategy, and explores the parameter space from multiple learning perspectives. Unlike the update strategy of traditional optimization methods which only rely on sampling of a single learning function, the proposed method combines multiple learning functions to achieve sample diversity improvement and modeling convergence acceleration in a multi-strategy collaborative manner, thereby reducing simulation call frequency and improving the response capability of modeling update.

[0007] A bridge digital twin modeling update system driven by cluster learning Kriging includes:

[0008] A data acquisition and processing module is used to acquire actual bridge structure response data under a preset working condition by using sensors deployed on a bridge in service, and to process performance characteristic information of the actual bridge structure.

[0009] A digital twin model construction module is used to construct a bridge digital twin model, simulate bridge structure response data under the preset working condition, and process performance characteristic information of the simulated bridge structure.

[0010] A target function construction module is used to construct a residual target function using performance characteristic information of the actual bridge structure and performance characteristic information of the simulated bridge structure, and to obtain optimization variables and parameter space of the optimization variables based on the residual target function.

[0011] An optimization module is used to guide Kriging surrogate model to search for optimal parameters in the parameter space by using a cluster optimization mechanism driven by multiple learning functions in parallel.

[0012] A model update module is used to update the bridge digital twin model through the optimal parameters, and to complete the construction and tracking of the bridge digital twin.

[0013] Preferably, in the data acquisition and processing module, the sensors used include vertical laser displacement meters and transverse vibration pickups, which are arranged at the midspan of the main girder; wherein the vertical laser displacement meters are used to measure deflection responses and assist in identifying vertical vibration characteristics; and the transverse vibration pickups are used to capture bridge transverse vibration characteristics.

[0014] Preferably, in the objective function construction module, the process of constructing the residual objective function includes:

[0015] extracting actual bridge natural frequencies from the actual bridge structure response data, and obtaining simulated bridge natural frequencies by using the bridge digital twin model;

[0016] Based on the actual bridge natural frequencies and the simulated bridge natural frequencies, taking the fundamental frequency of bridge transverse bending and vertical bending as the target frequency, the residual objective function is constructed.

[0017] Preferably, the optimization module includes:

[0018] An initial sample construction unit is configured to select an initial input sample point in the parameter space, calculate a response output corresponding to the initial input sample point, and construct an initial modeling sample;

[0019] A surrogate model construction unit is configured to construct an initial Kriging surrogate model based on the initial modeling sample;

[0020] A first optimization unit is configured to search for an optimal parameter in the parameter space, perform convergence evaluation on the optimal parameter, and if a preset convergence condition is met, output the current search result as a global optimal solution and terminate iteration; otherwise, execute a second optimization unit and enter the next round of optimization;

[0021] The second optimization unit is configured to construct an active learning function cluster according to the residual objective function, and based on different computing nodes, respectively perform a sub-optimization process on the active learning function cluster in parallel to obtain a plurality of optimal new input sample points;

[0022] A correlation evaluation unit is configured to evaluate the spatial correlation between the optimal new input sample points and between the optimal new input sample points and the initial input sample points, and delete redundant sample points whose spatial correlation does not meet a preset threshold;

[0023] A sample expansion unit is configured to calculate, in different computing nodes in parallel, response outputs corresponding to the optimal new input sample points after deleting the redundant sample points, obtain new added sample points, and expand them into the initial modeling sample;

[0024] An agent model updating unit is configured to update the initial Kriging agent model based on the expanded initial modeling sample and return the first optimization unit to continue iteration optimization until a preset convergence condition is met.

[0025] Preferably, in the initial sample construction unit, a Latin hypercube experimental design method is used to sample in the parameter space to obtain the initial input sample points.

[0026] Preferably, in the second optimization unit, the maximum mean square error function, the expected improvement function, the minimum agent model prediction function and the confidence lower bound function are used to construct the active learning function cluster.

[0027] The application further provides a bridge digital twin modeling updating method driven by cluster learning Kriging, and application of the system, comprising:

[0028] S1: Using sensors deployed on a bridge in service, actual bridge structure response data under a preset working condition is collected, and performance characteristic information of the actual bridge structure is obtained by processing;

[0029] S2: A bridge digital twin model is constructed to simulate bridge structure response data under the preset working condition, and performance characteristic information of the simulated bridge structure is obtained by processing;

[0030] S3: Using the performance characteristic information of the actual bridge structure and the performance characteristic information of the simulated bridge structure, a residual target function is constructed, and an optimization variable and a parameter space of the optimization variable are obtained based on the residual target function;

[0031] S4: A multi-learning function parallel driving cluster optimization mechanism is used to guide Kriging agent model to search for an optimal parameter in the parameter space;

[0032] S5: The bridge digital twin model is updated by the optimal parameter to complete construction and tracking of the bridge digital twin.

[0033] Preferably, in step S4, the method for searching for the optimal parameter in the parameter space comprises:

[0034] S41: An initial input sample point is selected in the parameter space, and a response output corresponding to the initial input sample point is calculated to construct an initial modeling sample;

[0035] S42: An initial Kriging agent model is constructed based on the initial modeling sample;

[0036] S43: search for optimal parameters in the parameter space, and perform convergence evaluation on the optimal parameters, if a preset convergence condition is met, the current search result is output as a global optimal solution, and iteration is terminated, otherwise, step S44 is performed, and next round optimization is entered;

[0037] S44: construct an active learning function cluster according to the residual target function, and perform sub-optimization processes on the active learning function cluster based on different computing nodes respectively to obtain a plurality of optimal new input sample points;

[0038] S45: evaluate the spatial correlation between the optimal new input sample points and the spatial correlation between the optimal new input sample points and the initial input sample points, and delete redundant sample points whose spatial correlation does not meet a preset threshold;

[0039] S46: parallel computing of the optimal new input sample points corresponding to the response output of the redundant sample points is performed in different computing nodes to obtain new sample points, and the new sample points are expanded into the initial modeling samples;

[0040] S47: the initial Kriging surrogate model is updated based on the expanded initial modeling samples, and step S43 is returned to continue iterative optimization until a preset convergence condition is met.

[0041] Compared with the prior art, the beneficial effects of the present application are:

[0042] The present application provides a cluster learning Kriging driven bridge digital twin modeling updating system and method, aiming to improve the precision of bridge structure state identification and the updating efficiency of digital twin model. The system is based on the response data of the actual bridge structure, and a virtual twin model corresponding thereto is constructed, and dynamic calibration and continuous updating of the model are realized through parameter optimization driving, so as to realize accurate modeling and evolution tracking of the bridge structure state.

[0043] The KCPL (multi-learning function parallel driving cluster optimization mechanism) algorithm proposed in the present application fuses a plurality of active learning functions, has multi-objective orientation and parallel sampling capability, significantly improves the optimization precision and convergence efficiency, and is suitable for parameter optimization under high-dimensional complex problems; the digital twin updating method based on the algorithm can realize efficient updating and accurate correction of the virtual model with lower computational overhead, breaks through the limitation that the traditional method is difficult to balance between response efficiency and modeling quality, improves the real-time performance and parameter identification precision of the bridge structure state identification, and has a wide application scenario.

[0044] The application not only improves the mapping accuracy of the bridge digital twin model to the real state of the structure, but also significantly enhances the performance in high-dimensional parameter identification, nonlinear response modeling and real-time updating. The constructed system has good scalability and engineering applicability, and is especially suitable for bridge digital twin application scenarios with significant dynamic evolution of the service state of the structure, high monitoring frequency and strict updating requirements. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed to be used in the embodiments. Obviously, the drawings described in the following embodiments are only some embodiments of the present application, and all other drawings obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0046] Figure 1 The structural schematic diagram of the cluster learning Kriging driven bridge digital twin modeling and updating system of the embodiment of the present application;

[0047] Figure 2 The framework flowchart of the cluster optimization mechanism (KCPL) driven by multiple learning functions in parallel of the embodiment of the present application;

[0048] Figure 3 The arrangement schematic diagram of the sensors in the data acquisition and processing module of the embodiment of the present application;

[0049] Figure 4 The frequency spectrum reflecting the natural vibration characteristics of the bridge structure of the embodiment of the present application; wherein (a) is the frequency spectrum of the lateral fundamental frequency; (b) is the frequency spectrum of the vertical fundamental frequency;

[0050] Figure 5 The schematic diagram of the bridge structure digital twin model of the embodiment of the present application;

[0051] Figure 6 The comparison diagram of the prediction results of different models and the measured data of the embodiment of the present application; wherein (a) is the response prediction comparison diagram at a speed of 40km / h; (b) is the response prediction comparison diagram at a speed of 60km / h; (c) is the response prediction comparison diagram at a speed of 70km / h; (d) is the response maximum value prediction comparison diagram. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0053] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0054] Embodiment one:

[0055] As shown in Figure 1 , a cluster learning Kriging driven bridge digital twin modeling updating system comprises a data acquisition and processing module, a digital twin model construction module, a target function construction module, an optimization module and a model updating module.

[0056] The data acquisition and processing module is used for acquiring actual bridge structure response data under a preset working condition by using sensors deployed on the in-service bridge, and processing to obtain performance characteristic information of the actual bridge structure. In a further embodiment, the sensors used in the data acquisition and processing module include vertical laser displacement meters and transverse vibration pickups, which are uniformly arranged at the midspan of the main girder. The vertical laser displacement meters are used to measure deflection response and assist in identifying vertical vibration characteristics, and the transverse vibration pickups are used to capture bridge transverse vibration characteristics. Figure 3

[0057] In the data acquisition and processing module, key indicators reflecting the structural performance characteristics are identified by processing the actual bridge structure response data. The key indicators are the natural frequencies of the bridge. As shown in Figure 4 , wherein, Figure 4 (a) is a spectrum diagram of the transverse fundamental frequency, Figure 4 (b) is a spectrum diagram of the vertical fundamental frequency; by performing spectrum analysis on the environmental excitation and the residual vibration response of the structure after the train passes, the main peak value in the spectrum density curve is extracted, and the dominant vibration frequency of the structure is identified; wherein, the main peak corresponding to the vertical bending of the U girder is located at 4.22 Hz, and the main peak of the transverse bending of the pier-girder system is located at 1.94 Hz. It can be known that the actual fundamental frequencies of the vertical bending of the U girder and the transverse bending of the pier-girder system are 4.22 Hz and 1.94 Hz respectively.

[0058] ​The digital twin model construction module is configured to construct a bridge digital twin model, simulate bridge structure response data under the preset working condition, and process performance characteristic information of the simulated bridge structure. Specifically, the digital twin model is based on a design file to select a four-node solid element SOLID45 to model the U-beam and the pier, and simultaneously consider the simulation of detailed structures such as the rail supporting platform and the bearing thickened section. The second-stage dead load is applied through a mass element MASS21. The material parameters used for modeling are the equivalent elastic modulus and the mass density. To improve the calculation accuracy and reduce the proportion of ill-conditioned elements, the main beam and the pier are divided into hexahedral elements with appropriate dimensions by means of volume sweeping. In terms of boundary conditions, the pier-beam connection is simulated by constraint coupling to simulate the simply supported constraint effect of the pier on the main beam, and the pier bottom is simulated by fixed constraint to reflect the support effect of the foundation on the pier. On this basis, the digital twin model is used to simulate the bridge structure response and the structural performance characteristic information under the preset working condition.

[0059] The objective function construction module is configured to construct a residual objective function by using the performance characteristic information of the actual bridge structure and the performance characteristic information of the simulated bridge structure, and obtain the optimization variables and the parameter space of the optimization variables based on the residual objective function. In a further embodiment, in the objective function construction module, the process of constructing the residual objective function includes extracting the actual bridge natural frequency from the actual bridge structure response data, and obtaining the simulated bridge natural frequency by using the bridge digital twin model. Based on the actual bridge natural frequency and the simulated bridge natural frequency, the residual objective function is constructed by taking the bridge lateral bending and vertical bending fundamental frequency as the target frequency to describe the deviation between them in the natural vibration characteristics. The residual objective function is as follows:

[0060]

[0061] In the formula, η is the parameter to be updated; f i (η) and f i,mea are the simulated frequency and the corresponding measured result, respectively; n is the number of target frequencies, and n=2 in this embodiment.

[0062] The sensitivity of each structural parameter to the dynamic characteristics of the bridge and the variability characteristics of the structural parameters are comprehensively considered to determine the optimization variables and the corresponding parameter space (i.e., the parameter variable design space) in the objective function. Specifically, the elastic modulus type parameters are allowed to float by ±30% relative to their design values, and the mass density type parameters are allowed to change by ±5%. Table 1 shows the specific values:

[0063] Table 1

[0064]

[0065] The optimization module is configured to guide the Kriging surrogate model to search for the optimal parameters in the parameter space by using the multi-learning function parallel driving cluster optimization mechanism. For example, Figure 2as shown.

[0066] Further embodiments are directed to the optimization module comprising:

[0067] An initial sample construction unit is configured to select initial input sample points in the parameter space and calculate the corresponding response outputs of the initial input sample points to construct initial modeling samples. In further embodiments, the initial sample construction unit uses a Latin hypercube experimental design method to sample in the parameter space to obtain the initial input sample points. Specifically, the Latin hypercube experimental design method divides each parameter dimension into 50 equidistant intervals and randomly samples in each interval, and then randomly recombines in the multi-dimensional space to generate 50 groups of initial sample points with good coverage to ensure uniform exploration of the parameter space. Based on the established bridge digital twin model, the response outputs corresponding to the input sample points are calculated numerically to form the initial modeling samples. In the numerical simulation process, the calculation tasks are distributed to four different calculation cores to accelerate the response outputs.

[0068] A proxy model construction unit is configured to construct an initial Kriging proxy model based on the initial modeling samples.

[0069] A first optimization unit is configured to search for optimal parameters in the parameter space and evaluate the convergence of the optimal parameters. If the preset convergence condition is met, the current search result is output as the global optimal solution and the iteration is terminated. Otherwise, a second optimization unit is executed to enter the next round of optimization. Specifically, a particle swarm optimization algorithm is used to search for the optimal parameter value by initializing a particle swarm, updating the particle position in each iteration according to the individual historical optimal value and the group optimal value, and evaluating the convergence. The convergence criteria include the following three items: any one of the target function value being lower than 1e-4, the number of iterations reaching 100, or the continuous improvement of the target function being less than 1e-6 for 30 times. If the evaluation result meets the set convergence condition, the current result can be output as the global optimal solution and the iteration is terminated. Otherwise, the second optimization unit is executed to enter the next round of optimization process.

[0070] A second optimization unit is configured to construct an active learning function cluster based on the residual target function and perform a sub-optimization process on the active learning function cluster in parallel based on different calculation nodes to obtain a plurality of optimal new input sample points. The active learning function cluster can be used to preliminarily obtain sample points for expansion, and the spatial correlation analysis can be used to evaluate the correlation between samples and delete redundant samples with high correlation to filter the sample points expanded in the previous step.

[0071] Further embodiments are that in the second optimization unit, a set of active learning functions, i.e. the multi-learning functions mentioned above, are constructed by using the maximum mean square error function (MSE), the expected improvement function (EI), the minimum surrogate model prediction function (MSP) and the lower confidence bound function (LCB). Based on different computing nodes, the learning function set is executed in parallel to obtain multiple optimal new input sample points. The functions and sampling criteria are as follows:

[0072] The expression of MSE is MSE(x) = σ(x), the characteristic description is the model mean square error estimate value, and the sampling criterion is x new = argmax[MSE(x)].

[0073] The expression of EI is The characteristic description is the balance trend of the current optimal value (minimum value) and the global search, and the sampling criterion is x new = argmax[EI(x)].

[0074] The expression of MSP is The characteristic description is the model prediction mean value, and the sampling criterion is x new = argmin[MSP(x)].

[0075] The expression of LCB is The characteristic description is the uncertainty penalty based on the model prediction, and the sampling criterion is x new = argmin[LCB(x)].

[0076] Note: y min is the current model optimal value (minimum value); Φ(·) and φ(·) are the standard normal cumulative distribution function and the probability density function, respectively.

[0077] The correlation evaluation unit is used to evaluate the spatial correlation between the optimal new input sample points and the initial input sample points, and to delete redundant sample points whose spatial correlation does not meet the preset threshold. When the correlation is too high, the spatial positions of two input points are particularly close or even coincide (when the spatial correlation is equal to 1, the two points completely coincide), and the contributions of the two points to subsequent optimization are almost the same. Therefore, for several points with particularly close spatial positions, only one point can obtain almost the same optimization effect as not deleting any point, so when the spatial correlation of different input samples is too high, it can be considered that there is no need to retain redundant points, and deleting them can save the subsequent optimization cost. In addition, Kriging modeling does not allow two input samples to be completely the same (the spatial correlation is equal to 1).

[0078] Specifically, the spatial correlation between new input sample points, new input sample points and initial input sample points is evaluated based on a Gaussian function, redundant samples with a correlation greater than 0.95 are deleted, and the Gaussian function expression is shown in formula (2):

[0079]

[0080] In the formula, θ l is a correlation function parameter, which can be solved by maximum likelihood estimation; is a component of different sample points x i , x j in the l dimension.

[0081] The sample expansion unit is used to calculate the response output corresponding to the optimal new input sample point for deleting redundant sample points in parallel in different computing nodes, obtain new point samples, and expand to the initial modeling samples.

[0082] The proxy model updating unit is used to update the initial Kriging proxy model based on the expanded initial modeling samples, and return to the first optimization unit to continue iterative optimization until the preset convergence condition is met.

[0083] Among them, in order to deeply analyze the accuracy and efficiency performance of the KCPL (multi-learning function parallel driven cluster optimization mechanism) algorithm proposed in the actual bridge twin update, the classical standard Kriging and PSO algorithm are introduced for comparison, and the related optimization parameters are set the same as the KCPL algorithm. Table 2 shows the structure transverse bending, vertical bending vibration frequency before updating and after updating by each method, and the error relative to the measured value. And table 3 summarizes the target function value, iteration number, modal analysis number and optimization time length and other core calculation indexes when each algorithm is updated, which are used to quantitatively analyze the performance of the proposed method.

[0084] Table 2

[0085]

[0086] Table 3

[0087] Computing metrics KCPL Kriging PSO Objective function value 7.90e-5 6.31e-4 9.61e-5 Total time (s) of run 551.48 273.13 1312.37 Total number of modal analyses 85 79 250

[0088] It can be seen that although the standard Kriging method has the highest calculation efficiency, it is difficult to finely adjust the key parameters due to the lack of active optimization mechanism, resulting in a significantly higher frequency fitting error than KCPL and PSO, and failing to meet the set convergence index. In contrast, the PSO method has higher fitting accuracy, but it is highly dependent on large-scale numerical simulation, and the calculation cost is significantly high. KCPL significantly reduces the calculation cost while maintaining comparable or even better fitting accuracy than PSO, fully embodying its good coordination ability in the contradiction between optimization accuracy and calculation cost in bridge digital twin modeling and updating.

[0089] a model updating module, used for updating the bridge digital twin model through the optimal parameters, completing the construction and tracking of the bridge digital twin. Figure 5 As shown in the figure. Specifically, the digital twin update is applied to the simulation prediction of the mid-span bottom plate vertical deflection when the train passes through the bridge at different speeds and the structural response under the train static loading condition.

[0090] Figure 6 The prediction results of the digital twin and the original design model under different working conditions and the comparison with the measured data are shown. Among them, Figure 6 (a) is a response prediction comparison chart under a vehicle speed of 40km / h; Figure 6 (b) is a response prediction comparison chart under a vehicle speed of 60km / h; Figure 6 (c) is a response prediction comparison chart under a vehicle speed of 70km / h; Figure 6 (d) is a response maximum prediction comparison chart. As Figure 6 Although both the digital twin and the virtual model based on the design file have reasonable mechanical expression ability, the original design model fails to reflect the actual service state and performance change of the structure, resulting in that it generally overestimates the real deformation of the bridge under the action of train load. In contrast, the prediction results of the digital twin model built based on the technical scheme of the present application are highly consistent with the measured data, which can more accurately capture the response trend, significantly improve the fitting peak value accuracy, and exhibit strong state mapping and prediction ability. The results further verify the identification accuracy of the present application to the current state of the structure, and reflect its practicability and promotion potential under complex operating conditions.

[0091] Example two:

[0092] The present application also provides a cluster learning Kriging driven bridge digital twin modeling updating method, and the system comprises:

[0093] S1: using the sensors deployed on the in-service bridge, collecting the actual bridge structure response data under the preset working condition, and processing to obtain the performance characteristic information of the actual bridge structure.

[0094] S2: constructing a bridge digital twin model, simulating the bridge structure response data under the preset working condition, and processing to obtain the performance characteristic information of the simulated bridge structure.

[0095] S3: using the performance characteristic information of the actual bridge structure and the performance characteristic information of the simulated bridge structure, constructing a residual target function, and obtaining the optimization variable and the parameter space of the optimization variable based on the residual target function.

[0096] S4: The cluster optimization mechanism driven in parallel by multiple learning functions guides the Kriging surrogate model to search for the optimal parameters in the parameter space.

[0097] S5: The bridge digital twin model is updated by the optimal parameters, and the construction and tracking of the bridge digital twin are completed.

[0098] Further embodiments are that in step S4, the method of searching for the optimal parameters in the parameter space comprises:

[0099] S41: Select initial input sample points in the parameter space, calculate the corresponding response outputs of the initial input sample points, and construct initial modeling samples.

[0100] S42: Based on the initial modeling samples, an initial Kriging surrogate model is constructed.

[0101] S43: Search for the optimal parameters in the parameter space, and perform convergence evaluation on the optimal parameters. If the preset convergence condition is met, the current search result is output as the global optimal solution, and the iteration is terminated; otherwise, step S44 is executed to enter the next round of optimization.

[0102] S44: Construct an active learning function cluster according to the residual error objective function, and perform a sub-optimization process on the active learning function cluster in parallel based on different computing nodes to obtain a plurality of optimal new input sample points.

[0103] S45: Evaluate the spatial correlation between the optimal new input sample points and the initial input sample points, and delete redundant sample points that do not meet the preset threshold.

[0104] S46: In different computing nodes, the response outputs corresponding to the optimal new input sample points after deleting the redundant sample points are calculated in parallel to obtain new added samples, which are expanded into the initial modeling samples.

[0105] S47: Update the initial Kriging surrogate model based on the expanded initial modeling samples, and return to step S43 to continue iterative optimization until the preset convergence condition is met.

[0106] The above-described embodiments are only descriptions of the preferred modes of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope of the present application as defined by the claims.

Claims

1. A swarm learning Kriging driven bridge digital twin modeling updating system, characterized in that, The method comprises the following steps: a data acquisition and processing module is used to acquire actual bridge structure response data under a preset working condition by using sensors deployed on a bridge in service, and to process performance characteristic information of the actual bridge structure; a digital twin model construction module is used to construct a bridge digital twin model, simulate bridge structure response data under the preset working condition, and process performance characteristic information of the simulated bridge structure; a target function construction module is used to construct a residual target function by using the performance characteristic information of the actual bridge structure and the performance characteristic information of the simulated bridge structure, and to obtain an optimization variable and a parameter space of the optimization variable based on the residual target function; an optimization module is used to guide a Kriging surrogate model to search for an optimal parameter in the parameter space by using a cluster optimization mechanism driven in parallel by multiple learning functions; a model updating module is used to update the bridge digital twin model by using the optimal parameter, and to complete construction and tracking of the bridge digital twin.

2. The system of claim 1, wherein, In the data acquisition and processing module, the sensors used include vertical laser displacement meters and transverse vibration pickups, which are uniformly arranged at the midspan of a main girder; the vertical laser displacement meters are used to measure deflection responses and assist in identifying vertical vibration characteristics; and the transverse vibration pickups are used to capture bridge transverse vibration characteristics.

3. The system of claim 1, wherein, In the target function construction module, the process of constructing the residual target function comprises the following steps: actual bridge natural frequencies are extracted from the actual bridge structure response data, and simulated bridge natural frequencies are obtained by using the bridge digital twin model; based on the actual bridge natural frequencies and the simulated bridge natural frequencies, the residual target function is constructed by taking bridge transverse bending and vertical bending fundamental frequencies as target frequencies.

4. The system of claim 1, wherein, The optimization module comprises the following units: an initial sample construction unit is used to select an initial input sample point in the parameter space, calculate a response output corresponding to the initial input sample point, and construct an initial modeling sample; a surrogate model construction unit is used to construct an initial Kriging surrogate model based on the initial modeling sample; a first optimization unit is used to search for an optimal parameter in the parameter space, perform convergence evaluation on the optimal parameter, and if a preset convergence condition is met, output the current search result as a global optimal solution and terminate iteration; otherwise, execute a second optimization unit and enter the next round of optimization; the second optimization unit is used to construct an active learning function cluster based on the residual target function, and based on different computing nodes, respectively perform a sub-optimization process on the active learning function cluster in parallel to obtain a plurality of optimal new input sample points; a correlation evaluation unit is used to evaluate spatial correlations between the optimal new input sample points and between the optimal new input sample points and the initial input sample point, and delete redundant sample points that do not satisfy a preset threshold in the spatial correlations; a sample expansion unit is used to calculate, in different computing nodes in parallel, response outputs corresponding to the optimal new input sample points from which the redundant sample points are deleted, obtain new point samples, and expand the new point samples into the initial modeling sample. An agent model updating unit is configured to update the initial Kriging agent model based on the expanded initial modeling samples and return to the first optimization unit for iterative optimization until a preset convergence condition is met.

5. The system of claim 4, wherein, In the initial sample construction unit, a Latin hypercube sampling method is used to sample in the parameter space to obtain the initial input sample points.

6. The system of claim 4, wherein, In the second optimization unit, the maximum mean square error function, the expected improvement function, the minimum agent model prediction function, and the confidence lower bound function are used to construct the active learning function cluster.

7. A swarm learning Kriging driven bridge digital twin modeling updating method, applying the system of any one of claims 1-6, characterized in that, The method comprises the following steps: S1: collecting actual bridge structure response data under a preset working condition by using sensors deployed on a bridge in service, and processing to obtain performance characteristic information of the actual bridge structure; S2: constructing a bridge digital twin model, simulating bridge structure response data under the preset working condition, and processing to obtain performance characteristic information of the simulated bridge structure; S3: constructing a residual objective function by using the performance characteristic information of the actual bridge structure and the performance characteristic information of the simulated bridge structure, and obtaining an optimization variable and a parameter space of the optimization variable based on the residual objective function; S4: using a cluster optimization mechanism driven in parallel by multiple learning functions to guide a Kriging agent model to search for an optimal parameter in the parameter space; S5: updating the bridge digital twin model by using the optimal parameter, and completing the construction and tracking of the bridge digital twin.

8. The method of claim 7, wherein, In step S4, the method for searching for the optimal parameter in the parameter space comprises: S41: selecting initial input sample points in the parameter space, calculating response outputs corresponding to the initial input sample points, and constructing initial modeling samples; S42: constructing an initial Kriging agent model based on the initial modeling samples; S43: searching for an optimal parameter in the parameter space, and performing convergence evaluation on the optimal parameter, if a preset convergence condition is met, the current search result is output as a global optimal solution, and the iteration is terminated; otherwise, step S44 is executed to enter the next round of optimization; S44: constructing an active learning function cluster according to the residual objective function, and performing a sub-optimization process on the active learning function cluster in parallel based on different computing nodes to obtain a plurality of optimal new input sample points; S45: evaluating the spatial correlation between the optimal new input sample points and the spatial correlation between the optimal new input sample points and the initial input sample points, and deleting redundant sample points that do not meet a preset threshold; S46: calculating the response outputs corresponding to the optimal new input sample points after the redundant sample points are deleted in parallel in different computing nodes to obtain new sample points, and expanding the new sample points into the initial modeling samples; S47: updating the initial Kriging agent model based on the expanded initial modeling samples, and returning to step S43 for iterative optimization until a preset convergence condition is met.

Images