Fragility curves establish a correlation between structural damage levels and seismic intensity,offering an intu-itive depiction of the probability of structural failure.However,the generation of these curves necessitates a sub-stantial amount of structural nonlinear time-history analysis results,thereby rendering the computational process in-efficient.Machine learning techniques have been proven to effectively address this issue,yet their efficacy dimini-shes with the increase in the scale of training data due to the computational demands of solving large-scale inverse matrices during the training phase.In response,this paper proposes a novel methodology,the Eigenvector Infor-mation-supported Support Vector Machine(EILS-SVM),which surmounts the limitations associated with these techniques.By employing a selective subsample to construct a low-rank kernel matrix in the context of large-scale datasets,the EILS-SVM method requires only the inversion of small-scale,low-rank matrices,significantly en-hancing computational efficiency.To validate the accuracy and efficiency of the EILS-SVM,it is benchmarked a-gainst conventional models such as the Least Squares Support Vector Machine(LS-SVM),Random Forest,Neu-ral Networks,Linear Discriminant Analysis(LDA),and Bayesian methods,using a dataset comprised of 16500 instances of damage in Reinforced Concrete(RC)frames subjected to seismic activities.The results indicate that the EILS-SVM is capable of accurately predicting the fragility curves of RC frames,with a computational efficiency improvement of up to 27 times compared to existing methodologies.
维普
万方数据