基于MLP-RF的框架结构残余变形预测研究OA

Residual deformation is an optimal parameter for evaluating structural damage after an earthquake.To investigate the relationship between ground motion parameters and structural residual deformation,this study focuses on frame structures.A large dataset of class-Ⅰ ground motions recorded on-site was selected,and ground motion parameters were calculated and then applied to finite element models established in ABAQUS for nonlinear time-history analysis,from which the converged structural residual deformations were extracted.By employing machine learning theory,an ensemble multilayer perceptron(MLP)-random forest(RF)learning model(MLP-RF)was devel-oped.The processed ground motion parameters and structural characteristics were used as inputs,with structural residual deformation as the output.MLP-RF was trained to predict the overall struc-tural residual deformation and interstory residual drift.Based on a comparison of prediction perfor-mance,the ground motion parameters exerting the greatest influence on residual deformation were identified.The results show that the prediction accuracy of MLP-RF is higher than that of the stand-alone MLP and RF models.The Newmark sliding displacement,Riddell displacement index,and mean acceleration response spectrum are the most influential ground motion parameters for residual deformation.The findings of this study provide invaluable insights into the underlying mechanism gov-erning the relationship between residual deformation and ground motion parameters.