Abstract:
|
In the design of tall structures, wind tunnel tests are crucial to ensuring safety as well as economy. The data produced from these tests are high frequency regular time series of pressures induced by wind at many locations on a model structure. Wind engineers are primarily concerned with very high and low pressures. The goal is to estimate the distribution of the highest and/or lowest pressure, over some length of time, usually longer than the actual wind tunnel test, for each location on the model structure. The peaks-over-threshold (POT) approach to this problem models all exceedances of a high threshold; however, all results are then conditional on the chosen threshold. In this poster, a POT approach to combining information across multiple thresholds is presented, and tested on thousands of real wind tunnel pressure series. The series are split into thirds, and the first one-third is used to predict the maximum of the remaining two-thirds. For these series, combining information from multiple thresholds out performs a single threshold, especially with respect to maintaining the nominal coverage of prediction intervals.
|