基于POT法确定风压系数极值的自动阈值选取与参数估计

Automated threshold selection and parameter estimation for determining extremewind pressure coefficients based on peaks over threshold method

风压系数极值是确定建筑围护结构设计风荷载的重要变量。实现阈值自动选取和合适的模型参数估计方法是保证超阈值模型极值计算结果精确性的先决条件,也是当前超越阈值模型研究的热点和难点。以CAARC高层建筑刚性模型测压风洞试验数据为基础开展超越阈值模型极值计算方法研究,通过对独立峰值数量和相关性研究独立峰值提取方法的性能;采用蒙特卡罗法研究4种不同的广义Pareto分布参数估计方法的性能,给出最佳参数估计方法选择建议;提出基于形状参数或极值估计结果稳定性的变点-局部比较阈值自动选取新方法。研究结果表明,基于变点理论-形状参数/极值稳定性阈值自动选取方法具有较小的样本依赖性,以及有较好的样本非高斯适用性,由此构建的改进超越阈值模型计算风压系数极值与标准极值的偏差小于5%,且完全实现阈值客观、自动选取,研究结论完善了小样本风压系数极值估计方法,对确定建筑围护结构设计风荷载具有重要意义,且可推广到其他极值估计领域。

The extreme pressure coefficient is an important variable to determine the design wind load of building’s claddings and components.The realization of automatic threshold selection and a proper parameter estimation method selection is a prerequisite to ensure the accuracy of the extreme value calculation results of the peaks over-threshold model,and it is the hotspot and difficulty of the peaks over-threshold model.Based on the wind tunnel tests of the rigid model of a CAARC high-rise building,the research on the extreme-pressure-coefficient calculation method about peaks over the threshold model was carried out,and the performance of the declustering method for independent peaks was studied by the number and correlation of independent peaks.The performance of the generalized Pareto distribution parameter estimation method was studied by the Monte Carlo approach,and the best parameter estimation method was proposed.A change point-local comparison method based on the stability of shape parameters or extreme value estimation results was proposed.The best threshold was automatically selected by the new method.The research results show that the automatic threshold selection method based on the change point theory has small sample dependence and good non-Gaussian applicability of samples.The proposed method can achieve a deviation of less than 5%,and fully realize automatic selection of threshold by the modified peak over threshold model.The research conclusions are helpful for extreme value estimation of pressure coefficients and are of great significance for determining design wind loads of building’s claddings and components,as well as can be extended to other extreme value estimation fields.