基于随机森林的中小跨径公路梁桥双柱墩地震易损性模型

Seismic fragility models of double column piers of small-and medium-span bridges in highway based on random forest

为准确建立线路中小跨径梁桥桥墩的地震易损性模型,采用统计工具得到某山区高速公路桥梁双柱墩的结构、几何和材料特性的概率分布,由拉丁超立方体抽样生成桥墩属性数据集,建立了参数化的有限元模型。通过Pushover分析和基于非弹性需求谱的能力谱方法获取双柱墩的地震需求和抗震性能数据点,提出由随机森林(RF)模型建立桥墩地震易损性的方法。结果表明:与传统方法相比,RF方法不对桥墩的地震需求和能力进行假设,避免了应用失实的构件能力值;RF模型具有较高的精度,各级IM上的精度均在83%以上;由RF模型中基尼指数的降低值可直接确定桥墩输入变量在易损性模型中的相对重要性,并且能够识别特定桥墩的损伤状态,避免重复的模拟计算。

In order to accurately establish the seismic fragility of twin column piers in a specific highway network,the probability distributions of structural,geometric,and material properties of the bridge portfolio in a mountainous area highway were obtained using statistical tools.The parameterized finite element models of the piers were established by pier sample sets generated from a Latin hypercube sampling method.The random forest(RF)models were adopted to establish the seismic vulnerability of piers applying both the earthquake demand and seismic performance data sets of the piers obtained by Pushover analyses and capacity spectrum method based on inelastic demand spectra,respectively.Results of the comparison of the proposed RF and traditional methods showed that the RF method didn't place any assumptions on the demand and capacity models of the piers,avoided the application of inaccurate component capacity values,and improved the accuracy with the value of greater than 83%.Also,the Gini index reductions of RF models can directly identify the relative importance of the uncertain input pier variables in the fragility models.The method can identify the damage states of the particular piers avoiding re-simulations.