Data-Driven Method for Probabilistic Response Prediction of Cable-Stayed Bridges

Abstract

This study proposes a data-driven method for predicting the probabilistic response of cable-stayed bridges. The proposed method is used to construct an optimal prediction model based on a data-driven machine-learning method. In addition, the accuracy and efficiency of the prediction model are improved by considering the correlation coefficients between the input sensor data and the output of the target response. The proposed method is comprised of two steps: the selection of meaningful features and the construction of a probabilistic prediction model employing Gaussian process regression. The proposed method is applied to an in-service cable-stayed bridge in the Republic of Korea using actual measurement data from various sensors. For comparison purposes, two parametric studies are performed, and the effects of the proposed feature-selection procedure are investigated based on the normalized correlation coefficients. Consequently, the proposed feature-selection method is proven to increase the accuracy and efficiency of the prediction.