Bridge deformation prediction based on SHM data using improved VMD and conditional KDE. (15th June 2022)

• Record Type: Journal Article
• Authors: Xin, Jingzhou; Jiang, Yan; Zhou, Jianting; Peng, Liuliu; Liu, Shuoyu; Tang, Qizhi
• Is Part Of: Engineering structures. Volume 261(2022)
• Publication Date: 2022-06-15
• Subjects: Bridge engineering -- Deformation prediction -- Structural health monitoring -- Improved variational mode decomposition -- Conditional kernel density estimation; Structural engineering -- Periodicals; Structural analysis (Engineering) -- Periodicals; Construction, Technique de la -- Périodiques; Génie parasismique -- Périodiques; Pression du vent -- Périodiques; Earthquake engineering; Structural engineering; Wind-pressure; Periodicals; 624.105
• DOI: 10.1016/j.engstruct.2022.114285
• Languages: English
• ISSNs: 0141-0296

Abstract

Highlights: A reliable, date-driven approach to predict bridge deformation using improved VMD and ARIMA-CKDE is proposed; An adaptive procedure in conjunction with KLD and GSO is developed to improve the performance of original VMD; Three groups of experimental data based on SHM system are used to conduct case study; A probabilistic prediction is given in the form of prediction interval with a 95% confidence level. Abstract: Deformation is a paramount index of bridge health monitoring. Accurate prediction of bridge deformation is of great significance to evaluate bridge performance. However, owing to the complex mechanical mechanism of bridge performance evolution, it is arduous to obtain a satisfactory forecasting result by simple data-driven methods. To this end, this paper proposes an innovative data-driven method based on an improved variational mode decomposition (IVMD) and conditional kernel density estimation (CKDE). Specifically, the raw deformation data are firstly pretreated by IVMD. This newly developed technique could not only adaptively optimize the decomposition level number through Hilbert transform and empirical mode decomposition, but also hinder the disturbance of irrelevant components by Kullback-Leibler divergence and Gram-Schmidt orthogonal. Then, auto-regression integrated moving average model is established to excavate the linear feature hidden in the data. Finally, CKDE in conjunction with Gaussian distribution assumption is designed to describe the above modeling errors, by which both point and interval predictions are generated. Obviously, this method avoids exploring the complex internal mechanism of structural behavior evolution. Three case studies based on structural health monitoring data are used to systematically evaluate its effectiveness. The results manifest that this method possesses higher reliability in comparison with the other concerned models, and could lay a foundation for early warning purpose.