Assessment of methods for extreme value analysis of non-Gaussian wind effects with short-term time history samples

Abstract

This study presents a comprehensive assessment of various methods for extreme value analysis of non-Gaussian wind effects using short-term time history samples. The methods examined are peaks-over-threshold (POT) method, the average conditional exceedance rate (ACER) method, and the translation process method with various translation models. The long-term wind pressure coefficient data on a saddle-shaped large-span roof collected from wind tunnel test are used as the basis for comparison. These pressure coefficient data are featured by a variety of non-Gaussian characteristics, including mildly and strongly softening and hardening non-Gaussian processes with unique distributions. Some new developments of the methods are also presented to better predict the extreme value distribution taking into account the non-Gaussian characteristics. The declustering of process to extract independent peaks over a given threshold for POT method is discussed. The effectiveness of the ACER method is firstly examined as applied to non-Gaussian wind pressures. Regarding the translation process method, this study highlights the limitations of widely used moment-based method and the method based on three-parameter gamma distribution of the process. A mixture distribution model is introduced for better modeling the distribution tail and estimation of extreme value distribution. This mixture distribution method and the method based on curve-fitting of translation function derived from mapping of cumulative distribution functions are illustrated to be capable of capturing the upper tail of translation function, thus lead to satisfactory estimations of extreme statistics for a variety of non-Gaussian processes.