Abstract: Due to the randomicity of extreme events like disaster and the otherness among data，it is often faced with the uncertainty of model when calculating risks by employing extreme value theory.To avoid the fitting error caused by improperly chosen of extreme value distributing model，the generalized extreme value distribution is adopted based on the introduction of related conception of extreme value theory.And the parameters of this complicated model are optimized by particle swarm method.Flight safety assessment is exemplified in simulation，and the results show that different distributing types can be optimized analogously，at the same time，ideal results can be achieved quickly，which may provide an effective approach for exactly selecting extreme value mode.

Key words: particle swarm algorithm, extreme value theory, generalized extreme value distribution, optimization

摘要： 由于灾难等极端事故发生的随机性以及数据之间的差异性，在运用极值理论计算风险时，必然面临模型的不确定性。为避免极值分布模型选择不当所引起的拟合误差，在介绍了极值理论相关概念的基础上，采用通用的广义极值分布模型，通过粒子群算法对复杂模型参数进行优化。以飞行安全为例进行仿真，实现了不同分布类型的统一优化处理，算法寻优速度较快、拟合效果理想，为准确选择极值分布模型提供了一条有效的途径。

关键词: 粒子群算法, 极值理论, 广义极值分布, 优化