Probabilistic model for multivariate non-stationary/non-Gaussian wind loads on the roof structure and potential engineering applications

The wind load field on the building surface under extreme winds contains multivariate non-stationary/non-Gaussian components, which is the direct reason for wind-induced structural destruction. Then, the probabilistic model for multivariate non-stationary/non-Gaussian multivariate wind loads is proposed. The mixture Gaussian distribution model and the appropriate copula are used to fit the marginal distribution and establish the dependence structures, respectively. Compared with the classical distribution model, the mixture Gaussian distribution model exhibits superior goodness of fitting, reducing the A-D values by 70 % and K-S values by 50 %. For selecting the optimal copula, plotting the frequency distribution histogram, and comparing the two tail dependence coefficients, the Chi-plotting method, the Akaike information criterion, and Bayesian information criterion principles are used. Based on the field measurement data under typhoon climate, the non-stationarity on the leeward side is stronger than that on the windward side. In determining the window length for non-stationary multivariate wind loads, the time-varying statistics are regarded as quantitative evaluation criteria, and the slope difference from the change point theory is used to identify the precise boundary point between different segments. For the synchronous multivariate non-stationary wind loads, the time-varying single copula model with evolutionary parameters and evolutionary multi-copula functions model with time are employed. The symmetrized Joe-Clayton copula model is applied in the probabilistic modeling of multivariate wind load fields on the corner region during the high wind speed interval of one typhoon.