Seismic Vulnerability Of RC Frame Structure Based On Machine Learning

Reinforced concrete(RC)frame structure is one of the typical structural systems in cities and towns in China.It is widely used in buildings that bear important social functions and is the key object of seismic performance design in China.Earthquake disasters in China over the years often cause damage to a large number of RC frame structures,which may lead to serious economic losses and a large number of casualties.In recent years,full probabilistic seismic engineering aiming at controlling seismic risk and seismic loss has become an important research trend in the field of seismic engineering.Seismic vulnerability analysis is a very important link,and seismic vulnerability assessment of building structures is an important work to ensure social security and stability.At present,the commonly used seismic vulnerability assessment method is computational analysis method.Its main work is to obtain the nonlinear relationship between ground motion intensity and structural response.This process requires a lot of elastoplastic time history analysis,which usually leads to the final assessment work cannot be completed efficiently.Therefore,how to avoid the time-consuming process in time-history analysis through other ways is one of the hot issues in the field of modern seismic engineering.Machine learning is a multi-disciplinary interdisciplinary subject that has emerged in recent years.It can deeply mine the complex relationship between feature data and target data.Based on this,in order to realize the accurate and efficient evaluation of the seismic vulnerability of RC frame structures,this paper studies and realizes the reasonable and efficient evaluation of the seismic vulnerability of three typical RC frame structures by introducing machine learning algorithms such as ensemble learning and neural network.The main contents of this study are as follows:(1)Typical RC frame structures with 3,8 and 14 floors are designed as examples.In order to reduce the influence of ground motion uncertainty and improve the reliability of sample data,based on the conditional mean spectrum(CMS)of the example site(Ya ’an area),the ground motion record is selected as the input,and the maximum inter-story displacement angle and the maximum horizontal displacement are used as the output.The sample database of structural response is built by incremental dynamic analysis(IDA)method.The ground motion intensity information and structural information are selected as the characteristic parameters,and the maximum inter-story displacement angle and the maximum horizontal displacement are used as the output parameters.(2)Based on two ensemble learning algorithms,extreme gradient boosting tree(XGBoost)and gradient boosting regression tree(GBRT),the reasonable prediction of structural response is studied.According to the generalization ability of the model,the two algorithms are selected,and the feature importance analysis is carried out to make the learning model easier to understand.(3)Based on artificial neural network(ANN),the reasonable prediction of structural response is realized,and the sensitivity analysis of characteristic parameters of artificial neural network is carried out by MIV analysis method.By comparing the generalization ability of artificial neural network and ensemble learning model,the machine learning algorithm most suitable for structural response prediction is selected.(4)Combining the feature analysis results of artificial neural network and ensemble learning algorithm,the feature parameters with the greatest contribution are selected.Based on the relationship between the selected characteristic parameters and the structural response output by the optimal machine learning algorithm,the probabilistic seismic demand model is obtained,and the rationality of the seismic vulnerability analysis method of RC frame structure based on machine learning is verified.Finally,the seismic vulnerability curve drawn by the proposed method is compared with the IDA calculation results to evaluate the accuracy of the output results of the new method.