基于仿真的复杂系统可靠性评估输入数据统计分析研究

Statistic Analysis of Input Data for Reliability Evaluation of Complex System with Simulation

对应用仿真技术评估复杂系统可靠性时输入数据的统计分析进行了深入研究。特别是对试验数据较少,即小样本情况进行了研究。对于根据小样本进行的可靠性估计,结合 Bayes 方法的蒙特卡洛(Bayes-MC)方法和结合改进的 Bootstrap 方法的蒙特卡洛(改进的 Bootstrap-MC)方法是比较有效的。概括总结了无数据情况下的专家经验估计三角分布方法,提出了改进的 Bootstrap 方法,将验前信息与专家经验纳入 Bootstrap 方法中,克服了该方法利用样本信息量不足的缺陷,使其更加完善与实用。在复杂系统的可靠性评估时,应采用综合或混合的方法。

Statistic analysis of input data is discussed for reliability evaluation of complex systems with simulation, especially when trial data is inadequate (i.e. small sample). Monte Carlo technique combined with Bayes method or improved Bootstrap method is appropriate for reliability evaluation in small sample cases. After discussion of triangular distribution method based on expert experience when no trial data are available, an improved Bootstrap method is proposed where Bootstrap method is combined with prior information and expert experience so that sample information can be well utilized. The integrated method is believed to be more practical in reliability evaluation of complex systems.