东南大学学报（自然科学版）2025，Vol.55Issue(6)：1522-1529,8.DOI:10.3969/j.issn.1001-0505.2025.06.004
基于循环神经网络辅助卡尔曼滤波法的动力响应重构方法
Dynamical response reconstruction method based on Kalman filter aided by recurrent neural network
摘要
Abstract
To address the problem that the traditional Kalman filter(KF)algorithm requires assuming both measurement noise and process noise as Gaussian white noise and necessitate manual specification of variances when reconstructing unmeasured structural responses,a dynamical response reconstruction method based on KF assisted by recurrent neural network was proposed.By integrating a gated recurrent unit neural network block into the recursive framework of the KF algorithm,the Kalman gain was calculated from the measured data,achieving state identification and response reconstruction and avoiding dependence from Gaussian noise distribution assumptions and prior knowledge.Numerical simulations and experimental analyses were con-ducted using a shear frame structure model.The results demonstrate that the proposed method can effectively achieve real-time response reconstruction.The average reconstruction errors of the displacement and accelera-tion in model experiments are 15.21%and 7.81%,respectively,exhibiting higher accuracy and robustness of the proposed method than those of the traditional KF algorithm.关键词
混合驱动建模/循环神经网络/卡尔曼滤波/响应重构Key words
hybrid modeling/recurrent neural network/Kalman filter/response reconstruction分类
土木建筑引用本文复制引用
孙利民,王艺晴,宋明明,夏烨..基于循环神经网络辅助卡尔曼滤波法的动力响应重构方法[J].东南大学学报（自然科学版）,2025,55(6):1522-1529,8.基金项目
国家自然科学基金资助项目(52208199,52378187). （52208199,52378187）
