基于自适应粒子滤波的结构损伤识别  

Structural Damage Identification Method Based on Adaptive Particle Filtering

作　　者：唐和生[1] 张伟[1] 陈镕[1] 薛松涛[1,2] 杨晓楠[1] 

机构地区：[1]同济大学结构工程与防灾研究所,上海200092 [2]日本近畿大学理工学部建筑学科,大阪577-8502

出　　处：《振动．测试与诊断》2008年第3期211-215,共5页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金资助项目(编号:50708076);教育部留学回国人员科研启动基金资助项目

摘　　要：为了解决传统粒子滤波方法粒子数量的选取较难确定的问题,提出一种非平稳动力系统突变参数识别的自适应粒子滤波方法(简称APF方法)。该方法利用系统后验概率密度与当前粒子集概率密度的K-L距离,自适应地更新采样粒子数量,在大幅降低识别过程中计算量的同时,不影响识别精度,使之更适合进行在线的结构系统参数识别。数值仿真结果发现,该方法的系统识别时间仅为传统粒子滤波方法的1/4,这证明了该方法在结构损伤在线识别中的有效性。For conventional particle filters the sample size is rather hard to determine for identifying a non-stationary dynam ic system with abrupt changes in system parameters. To solve this problem, an adaptive particle filtering (APF) method was propo sed. In the APF, the sampling size was updated according to the K-L distance between the s ystem posterior probability density and current probability density of sampling particles set, which decreased the computational load but did not affect its precision in the system identification process. Hence it would be more suitable for online tracking in structural parametric identification. Numerical simulation results show that the identification time required by APF is only a quarter of that by the conventional one, which proves the effectiveness of the proposed method in the online structural damage identification.

关 键 词：结构损伤识别 粒子滤波 K-L距离 自适应 

分 类 号：TU937.2[建筑科学—结构工程] TP274.2[自动化与计算机技术—检测技术与自动化装置]