A data-driven approach for modeling stochastic structural temperature field of a long-span cable-stayed bridge

Cyclic temperatures can induce significant structural effects on long-span cable-stayed bridges, challenging their safe operation. However, the thermal analysis of real bridges throughout their service life is computationally consuming, especially for long-span bridges. Therefore, it is essential to improve the computational efficiency for long-term structural temperature field simulation. This study proposes a simple-to-use data-driven approach for modeling the stochastic temperature field using monitoring data while considering time-varying temperature characteristics. Its key concept is that the temperature variances can be decomposed into periodic daily fluctuations and time-varying daily temperature extremes. The daily temperature extremes are modeled using Gaussian process regression and Copula theory, while their occurrence times are determined from the measured data using kernel density estimation. The approach is validated by comparing simulated and measured temperatures, focusing on both time histories and probability distributions. The results show a good agreement between the simulated and measured temperatures, confirming the validity of the proposed approach. • The long-term stochastic temperature field is directly modeled in the time domain with high efficiency. • The proposed approach is capable of modeling time-varying characteristics of temperatures across different seasons. • Temperature variations in the bridge are decomposed into daily extreme temperatures and periodic fluctuations. • The stochastic processes of daily extreme temperatures, considering spatial correlations, are derived. • A good agreement between the simulated and measured temperatures is found.