Simulating stochastic wind loads using spectral proper orthogonal decomposition

Monte Carlo-based methods are widely applied in reliability analysis of wind-excited structures, but require input from numerous realizations of stochastic wind loads or fields. Such realizations must preserve consistent second-order properties, including variance and power spectral density (PSD). Numerical simulation is a common substitute for repeating thousands of prohibitively expensive wind tunnel tests. The most popular simulation technique is the proper orthogonal decomposition-based method on the cross-spectral density matrix (XPOD). This study introduces a novel spectral proper orthogonal decomposition-based technique (SPOD) that significantly enhances simulation efficiency and accuracy. Results indicate that SPOD can significantly reduce the average relative error in PSD. Moreover, SPOD addresses the inherent shortcoming of XPOD in handling high-resolution spatial data. This study further identifies key parameters that critically influence the performance of both methods, especially input data segmentation. Based on the findings, practical recommendations for SPOD and XPOD simulations are given.