RC frame structure shock resistance toughness index prediction model construction and optimization method

Abstract

The invention discloses a method for constructing and optimizing an anti-seismic toughness index prediction model of an RC frame structure. The method comprises the following steps: determining structural design parameters and values thereof according to design requirements and structural basic information; determining a to-be-optimized target, namely a toughness index, according to design requirements, and performing structural toughness evaluation to obtain a value of the toughness index; the values of the structural design parameters and the values of the toughness indexes form a data set, and the data set is divided into a training set and a test set; by constructing a high-precision toughness index prediction model and combining an efficient multi-objective optimization algorithm, efficient and accurate optimization of RC frame structure design parameters is realized. A toughness index prediction model is embedded into a multi-objective optimization algorithm, and time-consuming numerical simulation is replaced by the toughness index prediction model, so that the multi-objective optimization process is accelerated. According to the optimization method, the technical problem that balance of structural design efficiency, economic cost and anti-seismic toughness indexes is difficult to consider in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the cross - field of RC frame structure section design and artificial intelligence, and specifically relates to a method for constructing and optimizing a prediction model of seismic resilience index for RC frame structures. Background Technique

[0002] In recent years, the application of machine learning (ML) technology and intelligent optimization algorithms has developed rapidly, attracting extensive research attention in the field of earthquake engineering. The ML method is easy to operate and can effectively predict output parameters with high precision according to the provided input parameters, significantly reducing the computational time cost required for the optimization design process. In addition, intelligent optimization algorithms can effectively handle the conflicts between multiple objective functions, with high search efficiency and convergence speed, ensuring accurate and efficient optimization results. Optimization algorithms have become important tools in various fields. This study completes the resilience assessment of RC frame structures based on machine learning and realizes the multi - objective optimization design of beam - column section sizes and reinforcement ratios through optimization algorithms, which is of great significance for promoting the development and progress of the field of structural earthquake resistance.

[0003] The traditional method for assessing the resilience of RC frame structures is based on vulnerability assessment, which requires repeating steps such as structural design, structural modeling, structural analysis, and structural parameter optimization, comparing the seismic reduction effects under different parameter combinations to determine the optimal structural design parameters. This method consumes a large amount of time cost and is difficult to balance multiple seismic indices, and it is impossible to obtain relevant resilience assessment results at the initial stage of structural design. The research on the seismic resilience assessment of existing building structures is still in the stage of continuous in - depth and improvement. Moreover, RC frame structures are a common type of structure in building structure design. Therefore, it is very necessary and urgent to carry out research on the method for seismic resilience assessment and section optimization design of RC structures based on machine learning.

[0004] In order to achieve the rapid prediction of structural resilience indices and the rapid optimization of structural design parameters, there is an urgent need for a method for constructing and optimizing a prediction model of seismic resilience indices for RC frame structures. Based on the OpenSees finite - element software, PACT resilience assessment software, and machine learning, a multi - objective optimization design framework for RC frame structures is established to achieve multi - objective optimization of the structural resilience index and cost. Overcome the problem that the traditional method has too high computational difficulty in solving multi - objective optimization problems, and realize the intelligence and precision of RC frame structure design. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a method for constructing and optimizing a prediction model of seismic resilience indices for RC frame structures, so as to solve the technical problems that the existing structural design method is difficult to balance multiple seismic indices and cannot obtain relevant resilience assessment results at the initial stage of structural design.

[0006] To solve the above technical problems, the present invention is implemented by the following technical solutions:

[0007] A method for constructing a prediction model of toughness index based on machine learning, specifically including the following steps:

[0008] Step 1, determine the structural design parameters and their values according to the design requirements and basic structural information; determine the target to be optimized according to the design requirements, that is, the toughness index, conduct structural toughness assessment, and obtain the values of the toughness index; form a data set with the values of the structural design parameters and the values of the toughness index, and divide the data set into a training set and a test set;

[0009] The structural design parameters include the storey height, span, and concrete strength grade of the RC frame structure, as well as the cross-sectional dimensions and reinforcement ratios of beams and columns;

[0010] The toughness indexes include the number of injured people, the number of dead people, and the repair cost;

[0011] Step 2, construct a machine learning prediction model, use the values of the structural design parameters in the training set as input parameters, use the values of the toughness indexes in the training set as output parameters, train the machine learning prediction model, and obtain the toughness index prediction model and its corresponding optimal hyperparameters;

[0012] The toughness index prediction model includes an injured number prediction model, a dead number prediction model, and a repair cost prediction model.

[0013] The present invention also includes the following technical features:

[0014] Step 1 specifically includes the following steps:

[0015] Step 1.1, determine the structural design parameters according to the design requirements and basic structural information; use the sampling method to take values for the structural design parameters to obtain n groups of structural design conditions, and each group of the structural design conditions includes a set of structural design parameter values;

[0016] Step 1.2, construct n finite element analysis models of the RC frame structure according to n groups of structural design conditions;

[0017] Step 1.3, use the LHS method to respectively match multiple seismic waves for each finite element analysis model for nonlinear time history analysis calculation, and a total of n groups of ground motion intensity indexes IM and structural loss indexes DM are obtained;

[0018] Step 1.4, obtain the median value of the structural collapse vulnerability according to each group of ground motion intensity indexes IM and structural loss indexes DM;

[0019] Step 1.5: Based on n sets of structural design working conditions and their corresponding median values of structural collapse vulnerability, perform ductility assessment on n finite element analysis models respectively to obtain the values of ductility indicators, that is, obtain a data set. The data set includes the values of structural design parameters and the values of ductility indicators, and divide the data set into a training set and a test set.

[0020] In Step 1.3, the ground motion intensity index IM includes peak acceleration; the structural loss index DM includes maximum inter-story drift ratio and maximum floor acceleration.

[0021] Step 2 specifically includes the following steps:

[0022] Step 2.1: Import the training set and the test set into the machine learning algorithm, and perform normalization processing on the values of the structural design parameters in the training set and the test set.

[0023] Step 2.2: Define hyperparameters and their search spaces.

[0024] Step 2.3: Randomly select three sets of hyperparameters in the search space, and use these three sets of hyperparameters to create three identical machine learning models respectively: the predicted number of injured people model, the predicted number of dead people model, and the predicted repair cost model, which together form the ductility index prediction model.

[0025] Step 2.4: Use the training set to train the predicted number of injured people model, the predicted number of dead people model, and the predicted repair cost model respectively to obtain the predicted number of injured people model, the predicted number of dead people model, and the predicted repair cost model after the current round of training, that is, the ductility index prediction model after the current round of training.

[0026] Step 2.5: Input the test set into the ductility index prediction model obtained from the current round of training, and output the predicted values of ductility indicators, evaluation indicators, and the values of hyperparameters; the predicted values of ductility indicators include the predicted values of the number of injured people, the number of dead people, and the repair cost.

[0027] Step 2.6: Return to Step 2.3 until i > N or t > T, and output the ductility index prediction model corresponding to the maximum value among all evaluation indicators and the corresponding values of the optimal hyperparameters.

[0028] i represents the number of trials;

[0029] N represents the maximum number of trials;

[0030] t represents the running time of the optimization process;

[0031] T represents the maximum running time of the optimization process.

[0032] In Step 3.4, the evaluation indicators include mean square error, root mean square error, and coefficient of determination.

[0033] An optimization method for the seismic resilience of an RC frame based on machine learning, comprising the following steps:

[0034] Step 1, construct a construction cost function f according to the material price of the RC frame structure c , the construction cost function f c and the repair cost prediction model obtained in the construction method of the resilience index prediction model based on machine learning to form a life cycle cost prediction model;

[0035] f c = A * B

[0036] A represents the material price;

[0037] B represents the material consumption;

[0038] Step 2, take the minimum value of the number of injured, the minimum value of the number of deaths, and the minimum value of the life cycle cost as the optimization objectives, and embed the injured number prediction model, the death number prediction model obtained in the construction method of the resilience index prediction model based on machine learning, and the life cycle cost prediction model obtained in Step 1 into the multi-objective optimization method to obtain the optimal solution set of the structural design parameters;

[0039] Step 3, determine the weight coefficients of each resilience index, and evaluate the optimal solution set through the following decision formula to obtain the optimal solution ROS of the structural design parameters corresponding to the current weight coefficients;

[0040]

[0041] In the formula:

[0042] n represents the number of solutions in the optimal solution set;

[0043] f 1max , f 1min represent the maximum and minimum values of the number of injured;

[0044] f 2max , f 2min represent the maximum and minimum values of the number of deaths;

[0045] f 3max and f 3min represent the maximum and minimum values of the life cycle cost;

[0046] ω1, ω2, ω3 represent the weight coefficients of the number of injured, the number of deaths, and the life cycle cost;

[0047] f 1i represents the number of injured corresponding to the i-th solution in the optimal solution set;

[0048] f 2i represents the number of deaths corresponding to the i-th solution in the optimal solution set;

[0049] f 3i represents the life cycle cost corresponding to the i-th solution in the optimal solution set.

[0050] 7. The method for optimizing the seismic resilience of an RC frame structure based on a machine learning-based RC framework according to claim 6, characterized in that, in step 2, the multi-objective optimization method includes a multi-objective particle swarm algorithm and a multi-objective differential evolution algorithm.

[0051] Compared with the prior art, the beneficial technical effects of the present invention are:

[0052] (Ⅰ) The present invention realizes the efficient and accurate optimization of the design parameters of the RC frame structure by constructing a high-precision resilience index prediction model and combining it with an efficient multi-objective optimization algorithm. By embedding the resilience index prediction model into the multi-objective optimization algorithm and replacing the time-consuming numerical simulation with the resilience index prediction model, the multi-objective optimization process is accelerated. The optimization method described in this paper overcomes the technical problem that it is difficult for the prior art to balance the structural design efficiency, economic cost and seismic resilience index.

[0053] (Ⅱ) In the construction of the resilience index prediction model, the present invention closely conforms to the development trend of the structural seismic design field from the traditional "collapse prevention" to "post-earthquake recoverability", and innovatively constructs a resilience index prediction model based on machine learning. Compared with the existing machine learning prediction models in the seismic field, this model unifies the structural safety assessment and seismic resilience analysis, and realizes the rapid and accurate assessment of the structural resilience index.

[0054] (Ⅲ) In the selection of the optimization objective, the present invention constructs a life cycle cost prediction model by combining the repair cost prediction model with the construction cost function, and uses it as the optimization objective. The life cycle cost can more accurately reflect the economic performance, help decision-makers optimize the project from a long-term perspective, and is more sustainable and economically beneficial.

[0055] (IV) In the determination of the optimal solution of the structural design parameters, the present invention constructs a decision formula to evaluate the optimal solution set. This quantitative formula comprehensively considers multiple optimization objectives such as the number of injured, the number of deaths and the life cycle cost, determines the weight coefficients of each objective according to the design requirements, and deals with the uncertainty factors in the evaluation process. It effectively solves the decision-making difficulty problem caused by the existence of multiple non-dominated solutions on the frontier of the optimal solution set in traditional multi-objective optimization design, and provides a scientific decision-making support tool for designers. Description of the Drawings

[0056] Figure 1 is the overall layout diagram of the RC frame structure;

[0057] Figure 2 It is the diagram of the Opensees finite element analysis model;

[0058] Figure 3 It is the frontier surface diagram of the optimal solution set of the structural design parameters.

[0059] The following further elaborates on the specific content of the present invention in conjunction with embodiments. Specific implementation manners

[0060] It should be noted that all components in the present invention, without special instructions, are components known in the art.

[0061] The following gives specific embodiments of the present invention. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent transformations made on the basis of the technical solutions of this application fall within the protection scope of the present invention.

[0062] The present invention provides a method for constructing a prediction model of toughness index based on machine learning, specifically including the following steps:

[0063] Step 1, determine the structural design parameters and their values according to the design requirements and basic structural information; determine the optimization target to be optimized according to the design requirements, that is, the toughness index, conduct structural toughness evaluation, and obtain the values of the toughness index; form a data set with the values of the structural design parameters and the values of the toughness index, and divide the data set into a training set and a test set;

[0064] The structural design parameters include the storey height, span, and concrete strength grade of the RC frame structure, as well as the cross-sectional dimensions and reinforcement ratios of beams and columns;

[0065] The toughness indices include the number of injured people, the number of dead people, and the repair cost;

[0066] Step 2, construct a machine learning prediction model, use the values of the structural design parameters in the training set as input parameters, use the values of the toughness index in the training set as output parameters, train the machine learning prediction model, and obtain the toughness index prediction model and its corresponding optimal hyperparameters;

[0067] The toughness index prediction model includes an injured people prediction model, a dead people prediction model, and a repair cost prediction model.

[0068] In the above technical solution, by constructing a high-precision prediction model for ductility index and combining with an efficient multi-objective optimization algorithm, the efficient and accurate optimization of the design parameters of the RC frame structure is realized. By embedding the ductility index prediction model into the multi-objective optimization algorithm and replacing the time-consuming numerical simulation with the ductility index prediction model, the multi-objective optimization process is accelerated. The optimization method in this paper overcomes the technical problem that it is difficult to balance the structural design efficiency, economic cost and seismic ductility index in the existing technology.

[0069] Step 1 specifically includes the following steps:

[0070] Step 1.1, determine the structural design parameters according to the design requirements and the basic structural information; use the sampling method to obtain the values of the structural design parameters, and obtain n groups of structural design conditions, where each group of structural design conditions includes a set of values of the structural design parameters;

[0071] Step 1.2, construct n finite element analysis models of the RC frame structure according to the n groups of structural design conditions;

[0072] Step 1.3, use the LHS method to match multiple seismic waves for each finite element analysis model respectively to perform nonlinear time history analysis calculations, and obtain n groups of ground motion intensity indexes IM and structural loss indexes DM in total;

[0073] Step 1.4, according to each group of ground motion intensity indexes IM and structural loss indexes DM, obtain the median value of the structural collapse vulnerability;

[0074] Step 1.5, based on the n groups of structural design conditions and their corresponding median values of the structural collapse vulnerability, perform ductility assessments on the n finite element analysis models respectively to obtain the values of the ductility indexes, that is, obtain the data set. The data set includes the values of the structural design parameters and the values of the ductility indexes, and divide the data set into a training set and a test set.

[0075] Preferably, the sampling method in Step 1.1 adopts Latin Hypercube Sampling (LHS); in Step 1.5, the ductility assessment can be completed by software such as PACT and SAUSG; in Step 1.2, finite element analysis software such as ABAQUS, Opensees, and Perform-3D is used to establish the finite element analysis model of the RC frame structure; in Step 1.3, the seismic waves can be obtained through channels such as the PEER website, the National Earthquake Data Center, and the open-source earthquake database.

[0076] In Step 1.3, the ground motion intensity index IM includes the peak acceleration; the structural loss index DM includes the maximum inter-story drift angle and the maximum floor acceleration.

[0077] Step 2 specifically includes the following steps:

[0078] Step 2.1: Import the training set and the test set into the machine learning algorithm, and perform normalization processing on the values of the structural design parameters in both the training set and the test set;

[0079] Step 2.2: Define the hyperparameters and their search spaces;

[0080] Step 2.3: Randomly select three groups of hyperparameters in the search space, and use these three groups of hyperparameters to create three identical machine learning models respectively: the injured number prediction model, the death number prediction model, and the repair cost prediction model, which together form the resilience index prediction model;

[0081] Step 2.4: Use the training set to train the injured number prediction model, the death number prediction model, and the repair cost prediction model respectively, and obtain the injured number prediction model, the death number prediction model, and the repair cost prediction model after the current round of training, that is, the resilience index prediction model after the current round of training;

[0082] Step 2.5: Input the test set into the resilience index prediction model obtained from the current round of training, and output the resilience index prediction values, evaluation indicators, and the values of the hyperparameters; the resilience index prediction values include the predicted values of the injured number, the death number, and the repair cost;

[0083] Step 2.6: Return to Step 2.3 until i > N or t > T, and output the resilience index prediction model corresponding to the maximum value among all evaluation indicators and the values of its corresponding optimal hyperparameters.

[0084] i represents the number of trials;

[0085] N represents the maximum number of trials;

[0086] t represents the running time of the optimization process;

[0087] T represents the maximum running time of the optimization process.

[0088] In the above technical solution, in terms of the construction of the resilience index prediction model, it closely conforms to the development trend of the structural seismic design field from the traditional "collapse prevention" to "post-earthquake recoverability", and innovatively constructs a resilience index prediction model based on machine learning. Compared with the existing machine learning prediction models in the seismic field, this model unifies the structural safety assessment and the seismic resilience analysis, and realizes the rapid and accurate assessment of the structural resilience index.

[0089] In Step 3.2, when defining the hyperparameters, the hyperparameter tuning algorithm can select grid search, random search, Bayesian optimization, or non-dominated sorting genetic algorithm;

[0090] In Step 3.4, the evaluation indicators include mean square error, root mean square error, and coefficient of determination.

[0091] The present invention provides an optimization method for the seismic resilience of an RC frame based on machine learning, which is characterized by the following steps:

[0092] Step 1, construct a construction cost function f according to the material price of the RC frame structure c , the construction cost function f c and the repair cost prediction model obtained in the construction method of the resilience index prediction model based on machine learning form a life cycle cost prediction model;

[0093] f c = A * B

[0094] A represents the material price;

[0095] B represents the material consumption;

[0096] Step 2, take the minimum value of the number of injured people, the minimum value of the number of deaths, and the minimum value of the life cycle cost as the optimization objectives, and embed the injured people prediction model, the death toll prediction model obtained in the construction method of the resilience index prediction model based on machine learning and the life cycle cost prediction model obtained in step one into the multi-objective optimization method to obtain the optimal solution set of the structural design parameters;

[0097] Step 3, determine the weight coefficients of each resilience index, evaluate the optimal solution set through the following decision formula, and obtain the optimal solution ROS of the structural design parameters corresponding to the current weight coefficients;

[0098]

[0099] In the formula:

[0100] n represents the number of solutions in the optimal solution set;

[0101] f 1max , f 1min represent the maximum and minimum values of the number of injured people;

[0102] f 2max , f 2min represent the maximum and minimum values of the number of deaths;

[0103] f 3max and f 3min represent the maximum and minimum values of the life cycle cost;

[0104] ω1, ω2, ω3 represent the weight coefficients of the number of injured people, the number of deaths, and the life cycle cost;

[0105] f 1i represents the number of injured people corresponding to the i-th solution in the optimal solution set;

[0106] f 2i represents the number of deaths corresponding to the \(i\)-th solution in the optimal solution set;

[0107] f 3i represents the life-cycle cost corresponding to the \(i\)-th solution in the optimal solution set.

[0108] In the above technical solution, by combining the repair cost prediction model with the construction cost function, a life-cycle cost prediction model is constructed and used as the optimization objective. The life-cycle cost can more accurately reflect the economic performance, help decision-makers optimize the project from a long-term perspective, and is more sustainable and economically beneficial. At the same time, a decision-making formula is constructed to evaluate the optimal solution set. This quantitative formula comprehensively considers multiple optimization objectives such as the number of injured, the number of deaths, and the life-cycle cost, determines the weight coefficients of each objective according to the design requirements, and deals with the uncertainty factors in the evaluation process, effectively solving the decision-making difficulty problem caused by the existence of multiple non-dominated solutions on the front of the optimal solution set in traditional multi-objective optimization design, and providing a scientific decision-making support tool for designers.

[0109] In the above technical solution, the value of the weight coefficient of each toughness index is given according to the requirements

[0110] In step 2, the multi-objective optimization method includes the multi-objective particle swarm optimization algorithm and the multi-objective differential evolution algorithm.

[0111] Embodiment:

[0112] In this embodiment, a five-story three-span RC frame structure is taken as an example, as Figure 1 shown. Earthquakes will cause varying degrees of damage to the structure, resulting in economic losses and casualties. Therefore, it is necessary to optimize the design of the structure to control the post-earthquake losses. In view of this, the construction and optimization method of the seismic toughness index prediction model for the RC frame structure is described, which specifically includes the following steps:

[0113] Step 1: Taking the RC frame structure as shown in Figure 1 as an example, the structural design parameters include the cross-sectional width of the frame beam (\(x_1\)), the cross-sectional height 1 of the frame beam (\(x_2\)), the cross-sectional height 2 of the frame beam (\(x_3\)), the longitudinal reinforcement ratio 1 of the frame beam (\(x_4\)), the longitudinal reinforcement ratio 2 of the frame beam (\(x_5\)), the mid-span reinforcement ratio of the frame beam (\(x_6\)), the span of the frame beam (\(x_7\)), the side length 1 of the frame column (\(x_8\)), the side length 2 of the frame column (\(x_9\)), the longitudinal reinforcement ratio 1 of the frame column (\(x 10 ), the longitudinal reinforcement ratio 2 of the frame column (\(x 11 ), the concrete strength grade (\(x 12 ), the floor height (\(x 13 );

[0114] Step 2: Determine the targets to be optimized (i.e., resilience indicators) as the number of injured personnel (y1), the number of deceased personnel (y2), and the repair cost (y3).

[0115] Step 3: According to the structural design parameters, use the Latin Hypercube Sampling (LHS) method to obtain values for each structural design parameter, resulting in 100 sets of structural design conditions, as shown in Table 1.

[0116] Table 1 Design of LHS parameter sample values

[0117]

[0118] Step 4: Based on the 100 sets of structural design conditions obtained in Step 3, use the Opensees finite element analysis platform to establish 100 sets of finite element analysis models, as Figure 2 shown.

[0119] Step 5: Obtain 180 seismic waves from the ground motion database developed by Baker et al. Use the LHS method to match 10 seismic waves for each of the 100 finite element models for nonlinear time history analysis calculations. Select the peak acceleration as the ground motion intensity index IM, and extract the maximum inter-story drift angle and maximum floor acceleration of the structure as the structural loss index DM.

[0120] Step 6: Calculate the median collapse vulnerability of the 100 structures based on each set of ground motion intensity index IM and structural loss index DM.

[0121] Step 7: Based on the 100 sets of structural design parameters and the corresponding median collapse vulnerability values, use the PACT software to evaluate the resilience of the 100 structures, obtaining the number of injured, the number of deceased, and the repair cost of each structure as resilience indicators, resulting in a dataset, as shown in Table 2, where 90% is used as the training set and 10% is used as the test set.

[0122] Table 2 Resilience indicators corresponding to n groups of RC frame structure models

[0123]

[0124] Step 8: Based on the python platform, select two machine learning algorithms, Random Forest and CatBoost. Import the training set and test set into the machine learning algorithms, and perform normalization processing on the values of the structural design parameters in both the training set and the test set.

[0125] Step 9: Select the grid search, tree-structured Bayesian optimization algorithm, and non-dominated sorting genetic algorithm with an elite strategy as the hyperparameter tuning algorithms, and define the hyperparameters and their search spaces.

[0126] Step 10: Create machine learning models, including the injured number prediction model, the death toll prediction model, and the repair cost prediction model, using the current hyperparameters. These models together form the resilience index prediction model;

[0127] Step 11: Use the training set to train the injured number prediction model, the death toll prediction model, and the repair cost prediction model respectively, to obtain the injured number prediction model, the death toll prediction model, and the repair cost prediction model after the current round of training, that is, the resilience index prediction model after the current round of training;

[0128] Step 12: Input the test set into the resilience index prediction model obtained from the current round of training, and output the predicted values of the resilience index, evaluation indicators, and the values of hyperparameters; The predicted values of the resilience index include the predicted values of the number of injured, the number of deaths, and the repair cost;

[0129] Step 13: Return to Step 10 until i > 100 or t > 1000s, and output the resilience index prediction model corresponding to the maximum value among all evaluation indicators and the corresponding optimal hyperparameter values.

[0130] Step 14: Construct the construction cost function f according to the material prices of the RC frame structure c , the construction cost function f c and the repair cost prediction model form the life cycle cost prediction model;

[0131] f c = (((x1 * x2 * (x4 + x6) * x7 * 10 -3 * 76 + x1 * x3 * (x5 + x6) * x7 * 10 -3 * 114)

[0132] + (x8 * x8 * x 10 * x 13 * 48 + x9 * x9 * x 11 * x 13 * 72)) * 10 -8 * 7.85 * 3800

[0133] + (x1 * x2 * x7 * 10 -3 * 76 + x1 * x3 * x7 * 10 -3 * 114 + x8 * x8 * x 13 * 48

[0134] + x9 * x9 * x 13 * 72) * 10 -6 * 2360 * 1.4) / 7.19 * 6 / 0.6 * 2

[0135] Where: x1, x2, x3, ……, x nRepresents the values of the structural design parameters;

[0136] Step 15: Take the minimum values of the number of injured, the number of deaths, and the life - cycle cost as the optimization objectives, and embed the prediction models of the number of injured, the number of deaths, and the life - cycle cost into the multi - objective optimization method to obtain the optimal solution set of 50 groups of RC frame structural design parameters. The front - edge surface of the optimal solution set of the fitted structural design parameters is as Figure 3 shown;

[0137] Step 16: Take the weight coefficients of personnel injury, personnel death, and life - cycle cost as 0.45, 0.45, and 0.1 respectively. Evaluate the optimal solution set through the decision formula, and the optimal solution of the structural design parameters is shown in Table 3, and the comparison of the optimization effects is shown in Table 4;

[0138] Table 3 Relative optimal solutions of structural design parameters

[0139]

[0140] Table 4 Comparison of optimization effects

[0141]

Claims

1. A method for constructing a prediction model of toughness index based on machine learning, characterized in that Specifically, it includes the following steps: Step 1: Determine the structural design parameters and their values according to the design requirements and basic structural information; determine the optimization target to be optimized according to the design requirements, that is, the toughness index, conduct a structural toughness assessment, and obtain the value of the toughness index; Form a data set with the values of the structural design parameters and the values of the toughness index, and divide the data set into a training set and a test set; The structural design parameters include the storey height, span and concrete strength grade of the RC frame structure, as well as the cross-sectional dimensions and reinforcement ratios of beams and columns; The toughness index includes the number of injured people, the number of dead people and the repair cost; Step 2: Build a machine learning prediction model, use the values of the structural design parameters in the training set as input parameters, use the values of the toughness index in the training set as output parameters, train the machine learning prediction model, and obtain the toughness index prediction model and its corresponding optimal hyperparameters; The toughness index prediction model includes an injured people prediction model, a dead people prediction model and a repair cost prediction model.

2. The method for constructing a prediction model of toughness index based on machine learning according to claim 1, wherein, Step 1 specifically includes the following steps: Step 1.1: Determine the structural design parameters according to the design requirements and basic structural information; use the sampling method to obtain the values of the structural design parameters, and obtain n groups of structural design conditions, and each group of the structural design conditions includes a set of structural design parameter values; Step 1.2: Build n finite element analysis models of the RC frame structure according to n groups of structural design conditions; Step 1.3: Use the LHS method to respectively match multiple seismic waves for each finite element analysis model for nonlinear time history analysis and calculation, and obtain n groups of ground motion intensity indices IM and structural loss indices DM in total; Step 1.4: Obtain the median value of the structural collapse vulnerability according to each group of ground motion intensity indices IM and structural loss indices DM; Step 1.5: Based on n groups of structural design conditions and their corresponding median values of structural collapse vulnerability, conduct a toughness assessment on n finite element analysis models respectively, and obtain the values of the toughness index respectively, that is, obtain a data set, and the data set includes the values of the structural design parameters and the values of the toughness index, and divide the data set into a training set and a test set.

3. The construction method of the toughness index prediction model based on machine learning according to claim 2, wherein In Step 1.3, the ground motion intensity index IM includes the peak acceleration; the structural loss index DM includes the maximum inter-storey drift angle and the maximum floor acceleration.

4. The method for constructing a toughness index prediction model based on machine learning according to claim 1, wherein Step 2 specifically includes the following steps: Step 2.1: Import the training set and the test set into the machine learning algorithm, and normalize the values of the structural design parameters in the training set and the test set; Step 2.2: Define the hyperparameters and their search space; Step 2.3: Randomly select three groups of hyperparameters in the search space, and use these three groups of hyperparameters to create three identical machine learning models respectively: an injured people prediction model, a dead people prediction model and a repair cost prediction model, and jointly form a toughness index prediction model; Step 2.4: Use the training set to train the injured people prediction model, the dead people prediction model and the repair cost prediction model respectively, and obtain the injured people prediction model, the dead people prediction model and the repair cost prediction model after the current round of training, that is, the toughness index prediction model after the current round of training; Step 2.5, input the test set into the toughness index prediction model obtained from the current round of training, and output the predicted values of toughness indices, evaluation metrics, and the values of hyperparameters; the predicted values of toughness indices include the predicted values of the number of injured, the number of deaths, and the repair cost. Step 2.6, return to Step 2.3 until i > N or t > T, and output the toughness index prediction model corresponding to the maximum value among all evaluation metrics and the corresponding values of the optimal hyperparameters. i represents the number of trials. N represents the maximum number of trials. t represents the running time of the optimization process. T represents the maximum running time of the optimization process.

5. The construction method of the toughness index prediction model based on machine learning according to claim 4, characterized in that, In Step 3.4, the evaluation metrics include mean squared error, root mean squared error, and coefficient of determination.

6. An optimization method for the structural seismic resilience of an RC framework based on machine learning, characterized in that, It includes the following steps: Step 1, construct a construction cost function f according to the material prices of the RC frame structure c , the construction cost function f c and the repair cost prediction model obtained in the construction method of the resilience index prediction model based on machine learning according to any one of claims 1 to 5 to form a life cycle cost prediction model; f c = A * B A represents the material price. B represents the material consumption. Step 2, take the minimum value of the number of injured, the minimum value of the number of deaths, and the minimum value of the life cycle cost as the optimization objectives, embed the predicted model of the number of injured, the predicted model of the number of deaths obtained from the construction method of the machine learning-based toughness index prediction model according to any one of Claims 1 to 5, and the life cycle cost prediction model obtained in Step 1 into the multi-objective optimization method to obtain the optimal solution set of the structural design parameters. Step 3, determine the weight coefficients of each toughness index (it is stated in the specification that the coefficients are determined according to needs), evaluate the optimal solution set through the following decision formula, and obtain the optimal solution ROS of the structural design parameters corresponding to the current weight coefficients. In the formula: n represents the number of solutions in the optimal solution set. f 1max ,f 1min represent the maximum and minimum values of the number of injured persons; f 2max ,f 2min represent the maximum and minimum values of the number of deaths; f 3max and f 3min represent the maximum and minimum values of the life cycle cost; ω1, ω2, ω3 represent the weight coefficients of the number of injured, the number of deaths, and the life cycle cost. f 1i represents the number of injured corresponding to the i-th solution in the optimal solution set; f 2i represents the number of deaths corresponding to the $i$-th solution in the optimal solution set; f 3i represents the life cycle cost corresponding to the i-th solution in the optimal solution set.

7. The optimization method for the seismic resilience of the structure of the machine learning-based RC framework according to claim 6, wherein In Step 2, the multi-objective optimization method includes multi-objective particle swarm algorithm and multi-objective differential evolution algorithm.

Images