Data-driven simulation of multivariate nonstationary winds: A hybrid multivariate empirical mode decomposition and spectral representation method

Abstract

Extreme winds such as thunderstorms and tornados typically exhibit nonstationary characteristics. Correspondingly, the structural response under these extreme winds tends to be nonstationary. To obtain more accurate nonstationary structural response, the time-domain analysis method is commonly used, which makes the simulation of multivariate nonstationary winds an essential prerequisite. Recently, some single sample-based simulation methods have received much attention due to their straightforwardness. In these schemes, the implied assumption of the random initial phase shift to represent the spatial correlation may not be always appropriate. In this study, a single sample-based simulation method which is a hybrid of the multivariate empirical mode decomposition (MEMD) and spectral representation method (SRM) is proposed. Central to this method is the MEMD-based IF spectral matrix used to naturally consider the spatial correlation of the simulated random process without any assumption and adopting SRM to generate the sample. This makes the proposed method more straightforward and realistic. Statistical characteristics of the simulated random process are presented in detail. Numerical examples are provided to demonstrate the effectiveness of the proposed method. Results show that the characteristic of the measured single sample can be effectively preserved in the simulated sample. The estimated covariance and cross-covariance also provide a good agreement with their target values. In addition, the proposed method is more convenient and realistic in maintaining the spatial correlation of the measured single sample as compared to the methods that invoke the assumption concerning the random initial phase shift. Accordingly, the proposed hybrid method offers an effective simulation of nonstationary random processes.