基于集成学习的冷成型钢结构地震易损性预测方法

Vulnerability Prediction Method of Cold-formed Steel Structure under Earthquakes Based on Ensemble Study

为了研究近场地震下冷成型钢结构的易损性预测方法,文章采用了以随机森林为代表的集成学习算法,以一栋多层冷成型钢结构模型的动力时程分析数据为基础,建立了近场地震的易损性预测模型,并将易损性预测结果和动力时程分析结果进行对比分析。结果表明:本文的模型精度指标R^(2)、NMSE和MAE分别为0.968、-1.8×10^(-5)和0.003,预测值与模拟值相差较小,表明选用随机森林算法预测冷成型钢结构近场地震易损性的方法可靠,且预测精度比较高。

In order to research the vulnerability prediction method of cold-formed steel structures under near-field earthquakes.In this paper,the ensemble learning algorithm represented by RandomForest is used to establish a vulnerability prediction model based on the dynamic time history analysis data of a multi-layer cold-formed steel structure model.The results of vulnerability prediction are compared with the results of dynamic time history analysis.The results show that:The accuracy indexes R^(2),NMSE and MAE of the model in this paper are 0.968,-1.8×10^(-5)and 0.003 respectively,which show that the differences between prediction values and the FEA values are rather small,indicating that Random Forest algorithm is reliable in predicting the near field seismic vulnerability of cold-formed steel structures,and the prediction accuracy is relatively high.