Research On The Prediction Model Of Amplification Effect And Nonlinear Effect Of Complex Sites Based On Machine Learning Algorithm

Many mega-city clusters and many important projects in China are located in complex sites such as high seismic intensity zones and sedimentary basins(river valley lands),which have ground vibration amplification effects and site nonlinear effects.In recent years,machine learning algorithms have been gradually applied in earthquake engineering,but there are fewer applications in site amplification effect and nonlinear effect.In this paper,we focus on the typical complex site type of sedimentary basin and carry out the research on the assessment of the amplification effect and nonlinear effect of ground shaking in complex sites based on machine learning,and the main research work and innovations are as follows:(1)The applicability of classical machine learning algorithms(artificial neural networks,support vector machines,random forests,and extreme learning machines)to the problem of predicting the amplification effect of ground shaking at complex sites is explored.Based on the indirect boundary element method to numerically solve for common local sites(canyons and river valleys),a dataset is established,and the effects of particle swarm and differential evolution algorithms on improving the accuracy of the prediction models are analyzed;then a proxy model for the amplification effect of ground shaking in two-dimensional canyons and two-dimensional sedimentary river valleys is constructed.The results show that the differential evolution-artificial neural network has the best prediction effect on the site effect;the correlation between the site effect and surface location and incident wave frequency is high,72% and 16%,respectively.(2)A three-dimensional sedimentary basin ground shaking amplification effect prediction model based on particle swarm-differential evolutionary-artificial neural network is developed.Based on the fast multipole boundary element method to solve the 3D sedimentary basin ground shaking response to accelerate the data set establishment,the proxy model is applied to the analysis of the basin effect in Tunisia,the analysis of the influence of the uncertainty of geotechnical medium parameters on the basin effect,and the analysis of the influence of the basin effect on the failure probability of typical building structures under seismic action.The results show that:the particle swarm-differential evolution algorithm can effectively alleviate the overfitting phenomenon in the training process of artificial neural networks;the use of the agent model for the study of parameter uncertainty in geotechnical media can significantly reduce the computational cost,and the parameter uncertainty causes significant variability in the ground shaking of sedimentary basins;the probability of structural failure in different locations of sedimentary basins under seismic effects may vary by 10%-50%.(3)An artificial neural network-based method was established to evaluate the nonlinear response of ground shaking in sedimentary basins.Based on the high-quality seismic record data,the mapping relationship between ground shaking parameters,site condition parameters and site nonlinear effects is constructed,the correctness of the empirical threshold value for discriminating the nonlinear response of sedimentary basins is verified,the sensitivity of the nonlinear effects of sedimentary basins to different combinations of ground shaking parameters and site condition parameters is studied,the optimal empirical model is given,and then the empirical model is used to realize the rapid assessment of site The optimal empirical model is given,and then the empirical model is used to achieve rapid assessment of nonlinear response of the site.The study shows that the correlation between various ground shaking conditions is high,while the correlation between site conditions is poor;the combination of proxy conditions proposed in this paper can improve the performance of artificial neural networks and help to reasonably evaluate the nonlinear characteristics of sedimentary basins.