Parameterized earthquake vulnerability model training method and post-earthquake axle network function evaluation method
Through the parametric seismic vulnerability model training method, the Gaussian process regression deep learning model is used to predict the mean value of bridge vulnerability, and is extended to the post-earthquake functional evaluation of the bridge network, which solves the lack of unified and unified bridge seismic damage prediction models in the existing technology. To solve the problem of insufficient bridge network assessment, it achieves efficient and accurate vulnerability prediction and assessment of bridges and bridge networks.
Patent Information
- Application Number
- CN202411542111.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-01-24
AI Technical Summary
The existing technology lacks a unified model in the prediction of bridge seismic damage, especially the lack of seismic damage assessment of bridge nets, which limits the development of highway bridge seismic damage prediction systems.
Using the parametric seismic vulnerability model training method, by obtaining bridge specific parameters, using uniform design tables, finite element analysis and nonlinear time history analysis, a Gaussian process regression deep learning model is constructed to predict the mean value of bridge vulnerability. And extended to the post-earthquake functional evaluation of bridge networks.
It achieves accurate and efficient prediction of the mean value of bridge vulnerability, improves the efficiency and accuracy of post-earthquake functional assessment of the bridge network, and can adapt to different machine learning technologies and application needs.
Smart Images

Figure CN119358344A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bridge engineering, and more particularly to a parameterized earthquake vulnerability model training method and a post-earthquake bridge network function evaluation method. Background Art
[0002] In recent decades, bridge engineering, as an important foundation for national development, has developed significantly. With the gradual advancement of national infrastructure construction, the number and span of new bridges have been greatly improved. At the same time, the relevant seismic specifications for bridge engineering are also gradually improving. However, bridge damage caused by frequent earthquakes is becoming more and more common. Once a bridge is damaged, the transportation network may be paralyzed; in earthquake rescue and disaster relief, bridges play the role of lifelines, and the consequences of their damage will be more serious.
[0003] The rational implementation of bridge seismic fortification needs to be based on a reasonable bridge earthquake damage prediction model. On the one hand, a reasonable bridge earthquake damage prediction model can make a relatively accurate assessment of earthquake losses in local areas from a macro perspective, including pre-earthquake assessment to control possible losses within an acceptable range, establish a reasonable earthquake early warning and monitoring mechanism, and quickly formulate rescue plans after the earthquake to provide support for government decision-making; on the other hand, the bridge earthquake damage prediction model can also provide a relatively accurate reference for the seismic design of specific bridges. Although my country has carried out a lot of research on the earthquake damage prediction model of highway bridges, the research is often only aimed at single bridges, and there is little research on the earthquake damage assessment of bridge networks. In addition, there is currently a lack of a relatively unified highway bridge earthquake damage prediction model, which to a certain extent restricts the development of the highway bridge earthquake damage prediction system. Summary of the invention
[0004] The purpose of the present invention is to provide a parameterized earthquake vulnerability model training method and a post-earthquake bridge network function evaluation method, which can achieve accurate and efficient prediction of the mean value of bridge vulnerability and evaluation of the post-earthquake function of the bridge network.
[0005] The present invention provides a parameterized earthquake vulnerability model training method, comprising the following steps: S1: obtaining bridge-specific parameters, and according to the bridge-specific parameters, obtaining bridge samples and median vulnerability using a uniform design table, a finite element analysis method, and a nonlinear time-history analysis method; S2: randomly dividing the bridge samples and the median vulnerability according to a preset ratio to obtain a training set and a validation set; S3: using the training set and the validation set to train and validate a Gaussian process regression deep learning model to obtain a parameterized earthquake vulnerability model.
[0006] Further, step S1 specifically includes: S11: obtaining bridge-specific parameters, and obtaining bridge samples according to the bridge-specific parameters by using a uniform design table; S12: obtaining a fragility median according to the bridge samples by using a finite element analysis method and a nonlinear time history analysis method; Furthermore, the above-mentioned bridge-specific parameters include bridge span, main beam area, longitudinal reinforcement ratio, pier height, pier cross-sectional size and number of piers.
[0007] Furthermore, the above preset ratio is 4:1.
[0008] Furthermore, step S3 specifically includes: S31: using the training set to train the Gaussian process regression deep learning model to obtain a trained Gaussian process regression deep learning model; S32: using the verification set to verify the trained Gaussian process regression deep learning model to obtain a parameterized seismic vulnerability model.
[0009] Furthermore, the above-mentioned Gaussian process regression deep learning model includes a kernel function, and the kernel function is a square exponential kernel.
[0010] Furthermore, step S32 specifically includes: S321: using the trained Gaussian process regression deep learning model to predict the verification set to obtain a prediction result; S322: based on the prediction result, performing a performance evaluation on the trained Gaussian process regression deep learning model to obtain a performance evaluation index value; when the performance evaluation index value is better than the preset index value, stop training to obtain a parameterized seismic vulnerability model; otherwise, return to step S2 and continue training.
[0011] The present invention provides a parameterized seismic vulnerability assessment method, comprising: using the above-mentioned parameterized seismic vulnerability model training method to obtain a parameterized seismic vulnerability model, and using the parameterized seismic vulnerability model to predict a bridge network to be assessed to obtain an assessment result.
[0012] The present invention provides a method for evaluating the function of a post-earthquake bridge network, comprising the following steps: obtaining an evaluation result according to specific parameters of each bridge in the bridge network to be evaluated using the above-mentioned parameterized earthquake vulnerability evaluation method; obtaining the damage probability of each bridge network according to specified earthquake motions using the evaluation result; and obtaining a bridge network service index according to the damage probability of each bridge network.
[0013] Furthermore, according to the damage probability of each bridge network, the bridge network service index is obtained, as shown in the formula: , in, For bridge network service indicators, is the total number of bridges in the network, For the The total capacity of each bridge, whose value is derived from the traffic distribution of the bridge network; is the residual function ratio of the bridge, which can be determined according to the damage state of the associated bridge The value of represents the remaining traffic capacity of the bridge after the earthquake.
[0014] The implementation of the parameterized earthquake vulnerability model training method and the post-earthquake bridge network function assessment method provided by the present invention has the following beneficial effects: The present invention constructs a training set according to a UD uniform design table, thereby improving the training efficiency while ensuring the uniformity of samples; the present invention adopts Gaussian Process Regression (GPR) to establish a seismic vulnerability model for ductile highway bridges, which can effectively simulate the complex and implicit relationship between the prediction vector (i.e., bridge design parameters) and the predicted vulnerability parameters. Compared with the traditional nonlinear time-history analysis of bridges, the model can more efficiently handle nonlinear relationships and uncertainties, and achieve accurate and efficient prediction of the mean value of bridge vulnerability; at the same time, the model is highly flexible, can adapt to different machine learning technologies, and can be adjusted according to specific application requirements; compared with the traditional bridges that take a long time to perform finite element analysis, the parameterized seismic vulnerability model of the present invention can obtain the mean value of vulnerability directly according to the bridge parameters without performing finite element analysis, which greatly improves the calculation efficiency of bridge vulnerability analysis; the bridge network function evaluation index of the present invention can calculate the bridge network service index based on the bridge failure probability and traffic flow to evaluate the post-seismic function of the bridge network. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a flow chart of the parameterized earthquake vulnerability model training method provided by the present invention; Figure 2 It is a schematic diagram of the implementation process of the parametric seismic vulnerability assessment method provided by the present invention; Figure 3 It is a training effect diagram of the GPR model provided by the present invention; Figure 4 It is a bridge network model with practical effect provided by the present invention; Figure 5 The present invention provides the ground vibration ( im =0.8g) after the bridge network damage state comparison diagram; Figure 6 The present invention provides the ground vibration ( im =0.8g) Bridge network service index comparison chart; Figure 7It is a flow chart of the post-earthquake bridge network function evaluation method provided by the present invention. DETAILED DESCRIPTION
[0016] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.
[0017] Figure 1 A schematic diagram of a parameterized earthquake vulnerability model training method of this embodiment is shown. In this embodiment, the parameterized earthquake vulnerability model training method includes the following steps: S1: obtaining bridge specific parameters, and obtaining bridge samples and median fragility values according to the bridge specific parameters by using a uniform design table, a finite element analysis method, and a nonlinear time history analysis method; Specifically, step S1 includes: S11: Obtaining bridge-specific parameters, and obtaining bridge samples using a uniform design table according to the bridge-specific parameters; Specifically, the bridge-specific parameters include bridge span, main beam area, longitudinal reinforcement ratio, pier height, pier cross-sectional dimensions and the number of piers; It should be noted that in step S11, a uniform design table is selected as part of the experimental design method when constructing bridge samples. Uniform Design (UD) is a standardized array table with uniformity constructed in imitation of an orthogonal table based on the application principle of number theory in multidimensional numerical integration, and is used for uniform experimental design. Using UD to generate bridge samples can make the experimental sample set more uniform, reduce the demand for the number of samples of the machine learning model, and improve the training efficiency of the machine learning model. Specifically, various parameters of the bridge are selected, and different sample combinations are formed using the UD uniform design table to generate uniform bridge samples. S12: according to the bridge sample, using a finite element analysis method and a nonlinear time history analysis method, obtaining a median fragility value; It should be noted that in step S12, an OpenSees (Open System for Earthquake Engineering Simulation) model is generated according to different bridge samples, a finite element model is established, and then a group of ground motions are selected to perform a nonlinear time history analysis (Nonlinear Time History Analysis) on each bridge sample to obtain the median vulnerability of the bridge; S2: randomly dividing the bridge samples and the median vulnerability according to a preset ratio to obtain a training set and a validation set; In an exemplary embodiment, the preset ratio is 4:1; S3: using the training set and the validation set to train and validate the Gaussian process regression deep learning model to obtain a parameterized earthquake vulnerability model; Specifically, in this embodiment, the bridge sample set is randomly divided into a training set and a validation set, and the GPR deep learning model is trained and validated to construct a parameterized seismic vulnerability model and a relationship between the median vulnerability and specific parameters of the bridge; Specifically, step S3 includes: S31: Using the training set to train the Gaussian process regression deep learning model to obtain a trained Gaussian process regression deep learning model; Specifically, the Gaussian process regression deep learning model includes a kernel function, and the kernel function is a square exponential kernel; It should be noted that the Gaussian Process Regression (GPR) method is used for modeling in the Gaussian Process Regression deep learning model. GPR has data-driven characteristics, can flexibly model complex systems, and accurately predict observed data. It works under the framework of probabilistic regression. The model uses GPR to establish a mapping relationship between the median vulnerability and specific parameters of the bridge, and uses the bridge sample data set as input and the median vulnerability as output for model training. Specifically, its input values are the bridge span, main beam area, longitudinal reinforcement ratio, pier height, pier cross-sectional size, and number of piers. There is no need to pre-set the number of model parameters during GPR training, and it can flexibly adapt to the complexity of the data to avoid overfitting or underfitting. It should be noted that when constructing the model, when the model adopts the optimal ground motion, i.e. the evaluation ground motion, the fragility dispersion of the bridge is considered to be constant because the fluctuation of the fragility dispersion is small, and the fragility mean is set as the only output value of the model. Therefore, the parameterized model constructs the relationship between the bridge parameters and the fragility mean. In a preferred embodiment, the kernel function of GPR uses a squared exponential kernel (SEkernel); S32: Using the verification set to verify the trained Gaussian process regression deep learning model to obtain a parameterized earthquake vulnerability model; Specifically, step S32 includes: S321: Use the trained Gaussian process regression deep learning model to predict the verification set to obtain a prediction result; S322: According to the prediction result, the performance of the trained Gaussian process regression deep learning model is evaluated to obtain a performance evaluation index value; when the performance evaluation index value is better than a preset index value, the training is stopped to obtain a parameterized seismic vulnerability model; otherwise, the training is returned to step S2 to continue; It should be noted that, in this embodiment, the above performance evaluation index value adopts the determination coefficient ( R ²), the above preset index value is 0.85, that is, when When , stop training; otherwise, return to step S2 and continue training; In another embodiment, the performance evaluation index value is expressed as mean square error ( MSE ), the above preset index value is 0.2, that is, when When , stop training; otherwise, return to step S2 and continue training; In another embodiment, the above performance evaluation index value adopts the root mean square error ( RMSE ), the above preset index value is 0.35, that is, when When , stop training; otherwise, return to step S2 and continue training; In a preferred embodiment, the above performance evaluation index value is simultaneously adopted using the determination coefficient ( R ²), mean square error ( MSE ) and the root mean square error ( RMSE ) to evaluate the effectiveness of the model. The preset indicator values corresponding to the three evaluation indicators are: , and ; Specifically, in the training process, this embodiment uses an iterative method, including repeatedly randomly dividing the data set into a training set and a validation set; using the trained model to make predictions on the validation set; evaluating the performance of the model on the validation set in each iteration; in this part, the mean square error ( MSE ), coefficient of determination ( R 2 ) and the root mean square error ( RMSE ) to evaluate the performance of the model; if the performance of the trained model does not meet the specific requirements of these indicators, retrain and revalidate until a well-trained model with satisfactory performance is found; In this embodiment, the parameterized seismic vulnerability model includes: a bridge sample data set, a GPR deep learning model; unlike previous vulnerability models that focus on finite element models (the existing technology still requires finite element analysis), the parameterized seismic vulnerability model of this embodiment can directly obtain the mean vulnerability of the bridge after inputting specific bridge parameters; the model is not limited to a single bridge, but can realize the vulnerability prediction of the bridge network, classify the damage, and evaluate its post-seismic performance, and is used to conduct post-seismic functional evaluation of the bridge network.
[0018] This embodiment provides a parameterized seismic vulnerability assessment method, comprising the following steps: using the above-mentioned parameterized seismic vulnerability model training method to obtain a parameterized seismic vulnerability model, using the parameterized seismic vulnerability model to predict a bridge network to be assessed, and obtaining an assessment result; Figure 2 It is a schematic diagram of the implementation process of the parametric seismic vulnerability assessment method provided by the present invention.
[0019] This embodiment provides a method for evaluating the function of a bridge network after an earthquake, comprising the following steps: According to the specific parameters of each bridge in the bridge network to be evaluated, the above-mentioned parametric seismic vulnerability assessment method is used to predict and obtain the assessment results; Specifically, the above-mentioned specific parameters include bridge span, main beam area, longitudinal reinforcement ratio, pier height, pier cross-sectional dimensions and number of piers; According to the designated earthquake motion, the damage probability of each bridge network is obtained by using the evaluation results; According to the damage probability of each bridge network, a bridge network service index is obtained; In a preferred embodiment, the bridge network service index is obtained according to the damage probability of each bridge network, as shown in the formula:
[0020] in, For bridge network service indicators, is the total number of bridges in the network, For the The total capacity of each bridge, whose value is derived from the traffic distribution of the bridge network; is the residual function ratio of the bridge, which can be determined according to the damage state of the associated bridge The value of represents the remaining traffic capacity of the bridge after the earthquake.
[0021] It should be noted that the above method uses the Bridge Network Service Indicators (BNSI) when evaluating the post-earthquake performance of the bridge network. It is defined as the ratio of the average traffic volume that the bridge network can withstand after an earthquake event to the total initial traffic volume. The Bridge Network Service Indicators can evaluate the post-earthquake function of the bridge network. BNSI The value of .
[0022] In some embodiments, the above parameterized earthquake vulnerability model training method can also be implemented in the following manner: Figure 2 The technical roadmap for parameterized seismic vulnerability assessment method includes the following steps: (1) Select various parameters of the bridge, use the UD uniform design table to form different sample combinations to generate uniform bridge samples, and generate OpenSees models based on different samples; then select a set of ground motions, perform nonlinear time history analysis on each bridge sample, and obtain the median fragility of the bridge; (2) Using bridge samples as input values and median fragility as output values, the dataset is randomly divided into training set and validation set in a ratio of 4:1 to train GPR; (3) During the training process, an iterative approach was used, including repeatedly randomly splitting the dataset into a training set and a validation set; using the trained model to make predictions on the validation set; and evaluating the performance of the model on the validation set in each iteration. In this part, the mean square error ( MSE ), coefficient of determination ( R 2 ) and the root mean square error ( RMSE ) to evaluate the performance of the model; if the performance of the trained model does not meet the specific requirements of these indicators, retrain and revalidate until a well-trained model with satisfactory performance is found, such as Figure 3 As shown; After obtaining the trained parameterized earthquake vulnerability model, the present invention uses an actual bridge network in Hubei Province as a practical case to illustrate the use of the model. Figure 4 The figure shows the bridge network used in the present invention. The model is used to evaluate the damage state after earthquake and compare it with nonlinear time history analysis. The results are shown in Figure 5 As shown, the error is within 10%.
[0023] In some embodiments, the above-mentioned post-earthquake bridge network function assessment method can also be implemented in the following manner, specifically including the following steps: determining specific parameters of each bridge in the bridge network (bridge span, main beam area, longitudinal reinforcement ratio, pier height, pier cross-sectional size and number of piers), using the model to predict the mean vulnerability of each bridge, then obtaining the probability of damage to each bridge network under a specified earthquake motion, and finally obtaining the bridge network service index to evaluate the post-earthquake function of the bridge network; Calculate on this basis BNSI The model's post-earthquake function assessment results for the bridge network are as follows: Figure 6 As shown, the figure shows the post-earthquake function evaluation results of the bridge network obtained by the invention and the nonlinear time history analysis respectively. The error value of 0.02 proves that the model can be used for post-earthquake bridge network function evaluation; Figure 7 It is a flow chart of the post-earthquake bridge network function evaluation method provided by the present invention. The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.
Claims
1. A parameterized earthquake vulnerability model training method, characterized in that: The following steps are involved: S1: obtaining bridge specific parameters, and obtaining bridge samples and median fragility values according to the bridge specific parameters by using a uniform design table, a finite element analysis method, and a nonlinear time history analysis method; S2: randomly dividing the bridge samples and the median vulnerability according to a preset ratio to obtain a training set and a validation set; S3: Using the training set and the validation set to train and validate the Gaussian process regression deep learning model to obtain a parameterized earthquake vulnerability model.
2. The parameterized earthquake vulnerability model training method according to claim 1, characterized in that: Step S1 specifically includes: S11: Obtaining bridge-specific parameters, and obtaining bridge samples using a uniform design table according to the bridge-specific parameters; S12: According to the bridge sample, a finite element analysis method and a nonlinear time history analysis method are used to obtain a median fragility value.
3. The parameterized earthquake vulnerability model training method according to claim 2, characterized in that: The bridge-specific parameters include bridge span, main beam area, longitudinal reinforcement ratio, pier height, pier cross-sectional dimensions and the number of piers.
4. The parameterized earthquake vulnerability model training method according to claim 1, characterized in that: The preset ratio is 4:
1.
5. The parameterized earthquake vulnerability model training method according to claim 1, characterized in that: Step S3 specifically includes: S31: Using the training set to train the Gaussian process regression deep learning model to obtain a trained Gaussian process regression deep learning model; S32: Using the verification set to verify the trained Gaussian process regression deep learning model to obtain a parameterized earthquake vulnerability model.
6. The parameterized earthquake vulnerability model training method according to claim 5, characterized in that: The Gaussian process regression deep learning model includes a kernel function, and the kernel function is a square exponential kernel.
7. The parameterized earthquake vulnerability model training method according to claim 5, characterized in that: Step S32 specifically includes: S321: Use the trained Gaussian process regression deep learning model to predict the verification set to obtain a prediction result; S322: According to the prediction results, the performance of the trained Gaussian process regression deep learning model is evaluated to obtain a performance evaluation index value; when the performance evaluation index value is better than the preset index value, the training is stopped to obtain a parameterized earthquake vulnerability model; otherwise, return to step S2 and continue training.
8. A parametric seismic vulnerability assessment method, characterized in that: include: A parameterized seismic vulnerability model is obtained by using the parameterized seismic vulnerability model training method as described in any one of claims 1 to 7, and the parameterized seismic vulnerability model is used to predict the bridge network to be evaluated to obtain an evaluation result.
9. A method for evaluating the function of a bridge network after an earthquake, characterized in that: The following steps are involved: According to the specific parameters of each bridge in the bridge network to be evaluated, the parametric seismic vulnerability assessment method as claimed in claim 8 is used to obtain an assessment result; According to the specified earthquake motion, the damage probability of each bridge network is obtained by using the evaluation results; According to the damage probability of each bridge network, a bridge network service index is obtained.
10. A method for evaluating the function of a bridge network after an earthquake, characterized in that: According to the damage probability of each bridge network, the bridge network service index is obtained, as shown in the formula: , in, For bridge network service indicators, is the total number of bridges in the network, For the The total capacity of each bridge, whose value is derived from the traffic distribution of the bridge network; is the residual function ratio of the bridge, which can be determined according to the damage state of the associated bridge The value of represents the remaining traffic capacity of the bridge after the earthquake.
Citation Information
Cited By
Bridge earthquake vulnerability prediction method and system based on machine learning
CN120597611A
