Research On Methods For Estimating Extreme Wind Loads On Roofs With Minority Samples

Probabilistic assessment of extreme wind loads on the structure has been one ofcritical issues in the wind engineering. According to the number of required samples,current procedures for estimating extreme values can be divided into two categories:One is based on a large number of samples and the classical extreme-value theory, theother is based on very few samples. The former can establish more precise probabilisticmodel of peak wind loads and obtain the extreme-value with a certain guaranteed rate.The latter requires the first to adopt certain measures, such as parent distribution ofsamples is assumed or independent peaks are extracted from samples, in order to obtainextreme estimates, but its accuracy will vary for different estimation methods andsample properties. Relatively speaking, the latter is more practical. Therefore it issignificant to conduct a systematic study on the latter method. The main content of thisstudy includes the following aspects:1. Based on the classical extreme-value theory and repeated wind tunnel tests,the precise probability model of extreme wind pressure on typical roof is proposed.To accessing the accuracy of the second category methods, repeated wind tunneltests were carried out for spherical roof and flat roof, and more than one thousandsamples of wind pressure process were obtained. The precise probability mode ofextreme wind pressure is obtained based on the classical extreme-value theory. Basedon the probability model, the probability characteristics of extreme wind pressure onroof is revealed and the extreme estimates of wind pressure are obtained to evaluateaccuracy of extreme-value estimate methods based on minority samples.2. Comparison of extreme-value estimate methods based on the assumption ofparent distribution is carried outExtreme-value estimate methods based on the assumption of parent distribution,such as the peak factor method and its improved methods, were used separately. Thestability and accuracy of these methods were investigated. The results indicate that, theestimated results of peak factor method is varied less, but most of values areunderestimated; the estimated results of Modified Hermite Model is varied greatly, andmost of values are overestimated; the estimated results of Sadek-Simiu method areclosest to exact value.3. Comparison of extreme-value estimate methods based on sampledindependent peaks is carried outExtreme-value estimate methods based on sample multi-independent peaks includePeak-Over-Threshold method (POT), r-LOS method and “method of IndependentStorms”(MIS). The stability and accuracy of these three methods were investigated.The results indicate that, the estimated results of MIS are varied less than the other two methods. Nevertheless, the estimated results of POT and r-LOS method are better thanthose of MIS in the region where skew and kurtosis coefficient fluctuate wildly.Considering the stableness and accuracy of the estimated results, r-LOS method isproposed in project.4. An extreme-value estimate method based on wind pressure pulse events isproposedA problem that time space between the adjacent peaks is too large to make peaknumber few, exists among current procedures for extracting independent peaks fromwind pressure process. To solve this problem, new procedure based on the physicalmeaning of "Wind Pressure Pulse events" is proposed to get independent peaks. Basedon this new procedure, improved methods for POT method, r-LOS method and MIS areobtained. The stability and accuracy of these methods is studied and the results indicatethat, the estimated results of improved methods are better than current methods.To solve the problem existing among extreme probability model of “method ofIndependent Storms”, an improved model is proposed; Combining proposed model andproposed peak-extracting method, an improved MIS is obtained. The stability andaccuracy of these methods is studied and the results show that, this improved MIS isbest among these improved methods of this study.