基于KmeansSMOTE的损伤诊断模型性能提升及在斜拉桥中的应用  

Performance Improvement of Damage Diagnosis Model Based on KmeansSMOTE and its Application in Cable-Stayed Bridge

作　　者：刘杰[1,2,3] 陈佳梦 Liu Jie;Chen Jiameng(State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;Hebei Key Laboratory of Diagnosis,Reconstruction and Anti-disaster of Civil Engineering,Zhangjiakou 075000,China;School of Civil Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)

机构地区：[1]石家庄铁道大学省部共建交通工程结构力学行为与系统安全国家重点实验室,河北石家庄050043 [2]河北省土木工程诊断、改造与抗灾重点实验室,河北张家口075000 [3]石家庄铁道大学土木工程学院,河北石家庄050043

出　　处：《石家庄铁道大学学报（自然科学版）》2023年第2期35-41,共7页Journal of Shijiazhuang Tiedao University(Natural Science Edition)

摘　　要：为解决损伤诊断数据集中损伤与未损伤类数据样本不平衡导致诊断模型预测结果不理想的问题,提出了一种诊断模型性能提升的新方法(V-KmeansSMOTE)。该方法在改进合成少数类过采样技术(KmeansSMOTE)方法的基础上结合方差过滤(variance filtering,VF)特征选择技术筛除位移、加速度等数据中的零方差特征并对筛选后的特征所对应的损伤类数据样本进行过采样处理,将该方法应用于一座H型斜拉桥作为数值算例。结果表明,支持向量机分类模型分类准确率、精确度和F1评分在数据集上平均提升分别为6.19%、7.93%和20.07%,K最近邻模型平均提升6.18%、7.23%和7.26%,验证了提出方法的有效性。In order to solve the problem that the imbalance between damaged and undamaged data samples in the damage diagnosis data set leading to unsatisfactory prediction results of the diagnosis model,a new method for improving the performance of the diagnosis model(V-KmeansSMOTE)was proposed.Based on the improved synthetic minority over-sampling technique(KmeansSMOTE)method,this method combined variance filtering(VF)feature selection technique to screen out the zero variance features in displacement,acceleration and other data,and over-sampled the damage data samples corresponding to the selected features.This method was applied to an H-type cable-stayed bridge as a numerical example.The results show that after the application of this method,the classification accuracy,precision and F1 score of the support vector machine classification model increases by 6.19%,7.93%and 20.07%on the data set,respectively,and the K nearest neighbor model increases by 6.18%,7.23%and 7.26%on average,which verifies the effectiveness of the proposed method.

关 键 词：斜拉桥损伤诊断 模型性能提升方法 过采样技术 特征选择 数据挖掘 

分 类 号：U448.27[建筑科学—桥梁与隧道工程]