Bootstrap-Based Techniques for Computing Confidence Intervals in Monte Carlo System Reliability Evaluation

摘要

Bootstrap methods are presented for constructing confidence intervals of reliability indices. The bootstrap is a general re-sampling procedure which can be used to estimate the sampling distribution of statistics, requiring few assumptions, little modeling and analysis and which can be applied in an automatic way. Statistical significance is evaluated based on empirical distributions generated from an observed sample, for example, through a Monte Carlo simulation. The approach is illustrated with reference to the evaluation of confidence intervals of reliability indices of different network systems. The examples show that confidence intervals for reliability indices can be determined without prior information on their distributions, with little additional computational effort, using a small underlying sample of data.