Predicting acceleration response of super‐tall buildings by support vector regression

Abstract

SummaryRecovering missing data of defective sensors is an important challenge for reliability of structural health monitoring systems and misjudgment of structural conditions. The present study concerns predicting corrupted data of lost sensors by support vector regression (SVR). The method is tuned via optimizing their parameters by observer–teacher–learner‐based optimization as a powerful meta‐heuristic algorithm. Their performances are compared in predicting the acceleration responses of two real‐world super‐tall buildings: Milad Tower, located in Tehran, and Canton Tower in Guangzhou. Also the minimum required of sensors to predict the acceleration responses are investigated. The results are evaluated by five statistical indices exhibiting that the optimized SVR has sufficient capacity to predict acceleration responses of both towers with limited number of sensors. The proposed method is of practical interest as it does not require finite element modeling of the structure to derive its dynamic responses.

Unsupervised structural damage identification based on covariance matrix and deep clustering

“…In machine learning methods, support vector machines are widely used in the field of structural health monitoring due to their excellent performance 13 . Rouzbeh Doroudi et al 14 use support vector regression to predict corrupted data from missing sensors. Cheng and Zheng 15 used least squares support vector machines (LS‐SVM) to extract latent variables from environmental variables for dam safety monitoring.…”