Machine learning based seismic design method of anti-falling beam limiting cable for cross-fault beam bridge

Abstract

The application discloses a kind of machine learning-based anti-seismic design method of anti-falling beam limiting cable of cross-fault beam bridge, which calculates the design category of anti-falling beam limiting cable of cross-fault beam bridge by using machine learning classification model, and then calculates the design stiffness of anti-falling beam limiting cable of cross-fault beam bridge by using machine learning regression model. The method designed by the application can simply, quickly and accurately limit the relative displacement of pier and beam within the allowable displacement, preventing the cross-fault beam bridge from falling beam or support damage. The design result of the application has good seismic performance, meets the displacement requirements of anti-falling beam under the dual action of seismic dynamic effect and permanent displacement caused by fault rupture, is beneficial to improve the safety of bridge under strong earthquake, and can be used in engineering.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of across fault beam bridge anti-falling beam limiting cable and seismic design method, belong to bridge seismic mitigation design field. BACKGROUND

[0002] Considering the topography, road planning, engineering cost, construction period and regional economic development and other constraints, building bridges on active faults, reducing the disaster loss caused by fault dislocation, has become an unavoidable problem in engineering construction.

[0003] Unlike conventional bridge seismic damage, the collapse of cross-fault bridge is mainly caused by the double action of seismic dynamic power and permanent displacement generated by fault rupture, and the seismic damage is the falling beam caused by large displacement of pier beam, which needs to be reduced by reasonable limiting seismic design of pier beam displacement, one solution is to limit the relative displacement of pier beam within the expected range by designing the stiffness of limiting cable. The anti-falling beam limiting cable seismic design method in the existing bridge seismic design specification does not consider the influence of permanent displacement, so it cannot guarantee the seismic safety of cross-fault bridge. In addition, an iterative design method of cross-fault beam bridge anti-falling beam limiting cable is proposed in domestic and foreign literatures, which can accurately calculate the design stiffness of cross-fault bridge anti-falling beam limiting cable, but the design process is complicated, which is not friendly to engineering designers, resulting in longer design time.

[0004] To solve the above problems, the present application provides a kind of cross-fault beam bridge anti-falling beam limiting cable seismic design method based on machine learning, which can efficiently and accurately calculate the required limiting cable design stiffness and protect the seismic safety of cross-fault bridge. SUMMARY

[0005] The purpose of the present application is to solve the problems of complex and time-consuming iterative design method of cross-fault bridge anti-falling beam limiting cable, and to provide a kind of cross-fault beam bridge anti-falling beam limiting cable seismic design method based on machine learning, the main purpose is to use machine learning classification model to calculate the design category of cross-fault beam bridge anti-falling beam limiting cable, and then use machine learning regression model to calculate the design stiffness of cross-fault beam bridge anti-falling beam limiting cable, to limit the relative displacement of cross-fault beam under seismic action within the allowable pier beam displacement simply, quickly and accurately.

[0006] The technical scheme adopted by the present application is: a kind of cross-fault beam bridge anti-falling beam limiting cable seismic design method based on machine learning, comprising the following steps:

[0007] S1, determine the preliminary design parameters of the cross-fault bridge, including the pier yield stiffness, bearing yield stiffness, pier displacement ductility, bearing maximum shear strain, bridge span, main beam mass, cap beam mass, allowable pier beam displacement, ground permanent displacement, structural first-order period and second-order period seismic response spectrum amplitude, etc.

[0008] S2, input the above preliminary design parameters into the machine learning classification model for the design of the cross-fault beam bridge anti-falling beam limiting cable, and determine the design category of the preliminary design parameters.

[0009] S21, if the output of the machine learning classification model is "type A: no need to design", it indicates that the cross-fault bridge under the current preliminary design parameters has no damage risk, and no anti-falling beam limiting cable design is needed;

[0010] S22, if the output of the machine learning classification model is "type C: initial parameters do not meet the design conditions", it indicates that the pier beam relative displacement cannot be constrained within the allowable pier beam displacement under the current preliminary design parameters, and the pier yield stiffness, bearing yield stiffness, pier displacement ductility, and bearing maximum shear strain need to be adjusted to form new design parameters, and then enter step S2;

[0011] S23, if the output of the machine learning classification model is "type B: can be designed", the preliminary design parameters are used as the design parameters, and the next step is entered.

[0012] S3, input the design parameters into the machine learning regression model for the design of the cross-fault beam bridge anti-falling beam limiting cable, and obtain the required anti-falling beam limiting cable stiffness value.

[0013] S4, considering the randomness of the machine learning model, the above steps S1-S3 are repeated 10 times, and the average value of all anti-falling beam limiting cable stiffness values is obtained as the final design value.

[0014] Further, in step S1, the ground permanent displacement, the structural first-order period and the second-order period seismic response spectrum amplitude can be obtained by the cross-fault ground motion frequency division simulation method. First, the possible earthquake magnitude of the cross-fault bridge engineering design site is determined according to historical earthquake data, fault survey data, etc. The geometric characteristics of the fault are determined by the magnitude-length-width formula. Second, the spatial random field theory is used to generate a finite fault slip distribution model, and the fault rupture velocity and slip rise time are determined according to historical earthquakes. Then, the random finite fault method is used to simulate the high-frequency component of the cross-fault ground motion, and the modified discrete wave number method is used to simulate the low-frequency component of the cross-fault ground motion. Finally, the cross-fault ground motion required for design is obtained by frequency superposition, and the ground permanent displacement, the structural first-order period and the second-order period seismic response spectrum amplitude are calculated.

[0015] Further, the machine learning classification model involved in the design of the cross-fault beam bridge anti-falling beam limiting cable in step S2 can be a support vector machine (SVM), a light gradient boosting machine (LGBM), an extreme gradient boosting (XGB), a CatBoost (CatB), an artificial neural network (ANN), etc. The machine learning model is trained by using the database generated by the iterative design method of the cross-fault beam bridge anti-falling beam limiting cable, and the database is balanced by using the synthetic minority over-sampling technique (SMOTE) technology.

[0016] Further, the parameter adjustment strategy in step S22 is to gradually reduce the pier yield stiffness, increase the pier displacement ductility, increase the support yield stiffness, and increase the maximum support shear.

[0017] Further, the machine learning regression model involved in the design of the cross-fault beam bridge anti-falling beam limiting cable in step S3 can be a random forest (RF), a light gradient boosting machine (LGBM), an extreme gradient boosting (XGB), a CatBoost (CatB), an artificial neural network (ANN), etc. The machine learning model is trained by using the data with designed limiting cable stiffness in the classification model database.

[0018] The beneficial effects of the present application are:

[0019] 1. The calculation efficiency is extremely high, and only 1 minute is needed for one calculation. The iterative design method of the cross-fault beam bridge anti-falling beam limiting cable needs several days to complete, so the design time cost is significantly saved.

[0020] 2. The calculation accuracy is high. The error of the classification and regression model of the present application is only 3% compared with the iterative design method, which can effectively constrain the relative displacement between the pier and beam of the cross-fault beam bridge within the allowable displacement.

[0021] 3. The design process is simple and easy to operate. The present application can give the limiting cable stiffness at one time, avoiding repeated iteration, and is friendly and easy to operate for design engineers. DETAILED DESCRIPTION

[0022] Figure 1 is the flow chart of the machine learning-based anti-seismic design method of the cross-fault beam bridge anti-falling beam limiting cable of the present application;

[0023] Figure 2 is the prediction result of the classification model of the present application;

[0024] Figure 3 is the prediction result of the regression model of the present application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present application clearer and more apparent, the present application will be further described in detail below in combination with the drawings and examples.

[0026] The present application provides a machine learning-based anti-seismic design method for a cross-fault beam bridge limiting cable, as shown in the formula (I): Figure 1 The method comprises the following steps:

[0027] S1, determining the preliminary design parameters of the cross-fault bridge, including the pier yield stiffness, support yield stiffness, pier displacement ductility, maximum shear strain of support, bridge span, main beam mass, cap beam mass, allowable pier beam displacement, ground permanent displacement, structural first-order period and second-order period seismic response spectrum amplitude, etc.

[0028] Among them, the pier yield stiffness, support yield stiffness, pier displacement ductility, maximum shear strain of support, bridge span, main beam mass, cap beam mass are determined according to the specific parameters of the bridge, and the ground permanent displacement, the structural first-order period and the second-order period seismic response spectrum amplitude can be obtained by the cross-fault ground motion frequency division simulation method. If the seismic magnitude of the bridge design engineering site is evaluated as 7, the rupture length L of the fault is about 42km and the width W is about 18km through the empirical prediction model of fault source parameters. The fault is divided into 21 along the length direction and 9 along the width direction, with a size of 2x2km sub-faults, respectively using the stochastic finite fault method to simulate the high-frequency component of cross-fault ground motion, using the modified discrete wave number method to simulate the low-frequency component of cross-fault ground motion, and finally through frequency superposition, the cross-fault ground motion required for design is obtained, and the ground permanent displacement is calculated as 544mm, and the structural first-order period and second-order period seismic displacement response spectrum is 581.5mm / s 2 and 581.5mm / s 2 .

[0029] S2, inputting the above preliminary design parameters into the machine learning classification model for the cross-fault beam bridge limiting cable design, judging the design category of the preliminary design parameters.

[0030] In this step, the method for establishing the machine learning classification model is as follows: according to the data distribution pattern and range in Table 1, 34337 groups of initial design data are sampled by Monte Carlo sampling, and the design results corresponding to all initial design parameters are obtained by using the iterative design method of the anti-falling beam limiting cable of the trans-fault beam bridge, among which 6250 groups of initial design parameters do not need to design the limiting cable, 9639 groups of initial design parameters cannot design the limiting cable, and 18448 groups of parameters need to design the limiting cable. The number of three types of data is balanced to 18448 groups of initial design parameters by using the synthetic minority over-sampling technique (SMOTE) technology, and the three types of data are labeled. Machine learning models such as support vector machine (SVM), light gradient boosting machine (LGBM), extreme gradient boosting (XGB), CatBoost (CatB), and artificial neural network (ANN) are used for classification training of the above parameters, and the optimal hyperparameters of each classification model are found by grid search method to obtain a machine learning classification model for high-precision anti-falling beam limiting cable design of trans-fault beam bridges. Table 2 lists the hyperparameters of all classification models. Figure 2 The confusion matrix of all classification models is shown, and it can be seen that the calculation accuracy of SVM is 94.5%, the calculation accuracy of LGBM is 95.7%, the calculation accuracy of XGB is 95.8%, the calculation accuracy of CatB is 96.4%, and the calculation accuracy of ANN is 97.3%.

[0031] Table 1: Distribution pattern and range of initial design parameters of trans-fault bridge

[0032]

[0033] Table 2: Hyperparameters of classification models

[0034]

[0035]

[0036] If the output of the machine learning classification model is "Type A: no need to design", it indicates that the trans-fault bridge under the current preliminary design parameters has no risk of damage, and no anti-falling beam limiting cable design is needed.

[0037] If the output of the machine learning classification model is "Type C: initial parameters do not meet the design conditions", it indicates that the current preliminary design parameters cannot design a limiting cable that can constrain the relative displacement of pier and beam within the allowable pier and beam displacement, and it is necessary to return to step S1 to adjust the pier yield stiffness, support yield stiffness, pier displacement ductility, and support maximum shear strain to form new design parameters, and then enter step S2. The parameter adjustment strategy is to gradually reduce the pier yield stiffness, increase the pier displacement ductility, increase the support yield stiffness, and increase the support maximum shear strain.

[0038] If the output of the machine learning classification model is "Type B: can be designed", the preliminary design parameters are taken as the design parameters, and the next step is entered.

[0039] S3, input the design parameters into the machine learning regression model for the design of the anti-falling beam limiting cable of the cross-fault beam bridge to obtain the required anti-falling beam limiting cable stiffness value.

[0040] In this step, the establishment method of the machine learning regression model is as follows: 18448 groups of parameters of the limiting cable to be designed are taken as training samples, and machine learning models such as random forest (RF), light gradient boosting machine (LGBM), extreme gradient boosting (XGB), CatBoost (CatB), and artificial neural network (ANN) are used for regression training, and the optimal hyperparameters of each regression model are found through grid search method, thereby obtaining a high-precision machine learning regression model for the design of the anti-falling beam limiting cable of the cross-fault beam bridge. The hyperparameters of all regression models are listed in Table 3. Figure 3 The prediction results of all regression models are shown, and it can be seen that the calculation accuracy of RF is 95%, the calculation accuracy of LGBM is 97%, the calculation accuracy of XGB is 96%, the calculation accuracy of CatB is 97%, and the calculation accuracy of ANN is 97%.

[0041] Table 3 Hyperparameters of Regression Models

[0042]

[0043] S4, considering the randomness of the machine learning model, the above S1-S3 steps are repeated 10 times, and the average value of all anti-falling beam limiting cable stiffness values is taken as the final design value.

[0044] The above calculation process is simple and easy to operate, and the calculation efficiency is extremely high. Once calculation only takes less than 1 minute, and repeated 10 times to obtain the final design value only takes 10 minutes; however, the iterative design method of the anti-falling beam limiting cable of the cross-fault beam bridge proposed in the domestic and foreign literature has very complex design steps, and needs to be iterated repeatedly to perform response spectrum calculation. For skilled engineering designers, it also takes 1 day to design, which highlights the advantages of the anti-seismic design method of the anti-falling beam limiting cable of the cross-fault beam bridge based on machine learning.

[0045] Through this embodiment, it is shown that the design method of the present application can calculate the correct anti-falling beam limiting cable stiffness, limit the relative displacement of the pier and beam of the cross-fault beam bridge within the allowable displacement, meet the displacement requirements of the anti-falling beam under the dual action of the dynamic action of the earthquake and the permanent displacement generated by the fault rupture, and improve the safety of the bridge under strong earthquakes, which can be used in engineering.

[0046] The method for anti-seismic design of the cross-fault beam bridge limiting cable based on machine learning is described in detail above, and the principle and implementation mode of the present application are described by applying specific examples. The above description of the examples is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation mode and application range. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A seismic design method for anti-falling beam restraint cables of cross-fault beam bridges based on machine learning, comprising the following steps: S1. Determine the preliminary design parameters of the bridge across the fault, including pier yield stiffness, bearing yield stiffness, pier displacement ductility, bearing maximum shear strain, bridge span, main beam mass, cap beam mass, allowable pier-beam displacement, ground permanent displacement, and seismic response spectrum amplitude of the first and second periods of the structure. S2. Input the above preliminary design parameters into the machine learning classification model for the anti-falling beam limit cable design of the cross-fault beam bridge to determine the design category of the preliminary design parameters. S21. If the output of the machine learning classification model is "Type A: No design required", it means that the bridge spanning the fault has no risk of damage under the current preliminary design parameters and does not need to be designed with anti-falling beam limiting cables. S22. If the output of the machine learning classification model is "Type C: Initial parameters do not meet design conditions", it means that under the current preliminary design parameters, it is impossible to design a limiting cable that can constrain the relative displacement of the pier and beam to within the allowable displacement of the pier and beam. It is necessary to return to step S1 to adjust the pier yield stiffness, support yield stiffness, pier displacement ductility, and support maximum shear strain to form new design parameters, and then enter step S2. S23. If the output of the machine learning classification model is "Type B: Can be designed", then use the preliminary design parameters as the design parameters and proceed to the next step. S3. Input the design parameters into the machine learning regression model for the anti-falling beam limiting cable design of the cross-fault beam bridge to obtain the required anti-falling beam limiting cable stiffness value. S4. Considering the randomness of the machine learning model, repeat steps S1-S3 above 10 times and obtain the average value of the stiffness of all anti-fall beam limiting cables as the final design value. In step S2, the machine learning classification model for the anti-falling beam limiting cable design of the cross-fault beam bridge is obtained by training a database generated by the iterative design method of the cross-fault beam bridge anti-falling beam limiting cable using machine learning models such as support vector machine, lightweight gradient boosting machine, extreme gradient boosting, CatBoost, and artificial neural network. The database is processed for data balance using synthetic minority class oversampling technology. In step S3, the machine learning regression model for the anti-falling limit cable design of the cross-fault beam bridge is obtained by training random forest, lightweight gradient booster, extreme gradient booster, CatBoost, artificial neural network and other machine learning models with data in the classification model database that have designed limit cable stiffness.

2. The seismic design method for anti-falling beam restraint cables of cross-fault beam bridges based on machine learning according to claim 1, characterized in that, In step S1, the ground permanent displacement, the first-order period of the structure, and the second-order period of the ground motion response spectrum amplitude are obtained through the cross-fault ground motion frequency division simulation method. First, the magnitude of the possible earthquakes at the design site of the cross-fault bridge project is determined based on historical earthquake data and fault survey data. The geometric characteristics of the fault are determined by the formula of magnitude to fault length and width. Second, a finite fault slip distribution model of the fault plane is generated using spatial random field theory, and the fault rupture velocity and slip rise time are determined based on historical earthquakes. Then, the high-frequency components of the cross-fault ground motion are simulated using the random finite fault method, and the low-frequency components of the cross-fault ground motion are simulated using the modified discrete wavenumber method. Finally, the cross-fault ground motion required for the design is obtained by frequency superposition, and the ground permanent displacement, the first-order period of the structure, and the second-order period of the ground motion response spectrum amplitude are calculated.

3. The seismic design method for anti-falling beam restraint cables of cross-fault beam bridges based on machine learning according to claim 1, characterized in that, In step S22, the parameter adjustment strategy is to gradually reduce the pier yield stiffness, increase the pier displacement ductility, increase the support yield stiffness, and increase the support maximum shear strain.

Images