Stochastic response of a cable-stayed bridge under non-stationary winds and waves using different surrogate models

The prediction of the structural stochastic response under strong winds and waves is important in the design of cross-sea bridges with longer spans, while the non-stationary characteristics of winds and waves may affect the accuracy. This paper investigates the effects of non-stationary winds and waves on the stochastic response of a cable-stayed bridge based on a wind-wave-bridge (WWB) system. In order to improve the computational efficiency, three surrogate models, i.e., the support vector regression (SVR), the BP neural network (BPN), and the Gaussian process regression (GPR), are established by correlating the environmental parameters with the bridge response. A comparison among the three models is conducted to find the optimal one, and the effects of the mean wind speed, the significant wave height, and the peak wave period on the bridge response are further investigated. The bridge responses of the stationary wind and wave fields are 0.05%-16% larger than those of the nonstationary wind and wave fields. Different surrogate models are applicable to different parts of the bridge. The SVR, BPN, and GPR models are recommended for predicting the response of the tower, the foundation, and the girder, respectively, and the sensitivity analysis reveals the effects of non-stationary winds and waves.