A maximum entropy model with fractional moments for probability density function estimation of wind pressures on low-rise building

Due to the fickle distribution characteristics, good approximation of the wind pressure probability density function for an entire roof is challenging, especially for tail region. Maximum entropy model (MEM) can theoretically generate the least biased distribution. However, classical MEM may lead to unreliable estimates of the corresponding integer moments. This study developed a new maximum entropy model with fractional moments. There are several important features: A new translation function is introduced, thus allowing negative data to be modeled. For a good bias-variance trade-off, the constraint number M was fixed as four. After estimating the initial value of the Lagrange multiplier lambda by a computationally simple linear equation system, a simple search was carried out to find a better solution. The generalized pattern search was adopted for determining the global optimal solution for the fractional moment orders alpha. The performance was benchmarked through typical field measurement data of wind pressures on the roof of a low-rise building under typhoon conditions. Compared with common models, the proposed method had better performance and more stable results for the dataset examined for the whole roof. This method is also beneficial for peak value evaluation and the simulation of wind pressures.

Automated summary

This study developed a new maximum entropy model with fractional moments to better approximate the wind pressure probability density function for an entire roof. By introducing a new translation function and adopting generalized pattern search, more accurate and stable results can be obtained.