基于Markov Chain Monte Carlo的幂律过程的Bayesian分析  

Bayesian analysis for the power law process based on Markov Chain Monte Carlo

作　　者：王燕萍[1] 吕震宙[1] 赵新攀[1] 

机构地区：[1]西北工业大学航空学院,西安710072

出　　处：《航空动力学报》2010年第1期152-159,共8页Journal of Aerospace Power

基　　金：国家自然科学基金(10572117;50875213);航空基金(2007ZA53012);新世纪优秀人才支持计划(NCET-05-0868);863计划(2007AA04Z401)

摘　　要：在多种合理的无信息先验分布下,基于Markov Chain Monte Carlo方法,提出了一种简单且易于抽样的幂律过程的Bayesian分析方法.所提方法将失效、时间截尾数据统一分析,能快捷地获取幂律过程模型参数的Markov Chain Monte Carlo样本,利用该样本不但能直接给出模型参数函数的后验分布,还能给出单样预测和双样预测的分析.一个经典工程数值算例说明了所提方法的可行性、合理性与有效性.该方法具有一定的优越性,可为小子样可靠性增长分析提供一种值得参考的方法.Based on Markov Chain Monte approach for Bayesian analysis of a power law Carlo （MCMC） technique, a simple sampling process was presented under various reasonable noninformative priors. The Bayesian approach provides a unified methodology for both time and failure truncated data. Markov Chain Monte Carlo samples for the power law process are easily obtained from the presented approach. Based on these MCMC samples, not only the posterior distributions of some parameter functions of the power law process are given directly, but also the methodologies for single-sample and two-sample prediction are given easily. The results from an engineering numerical example illustrate the feasibility, rationality and validity of the presented approach. The proposed approach has a certain degree of superiority, thus providing an alternative method for the reliability growth analysis of small-sized samples

关 键 词：Bayesian推断 幂律过程 单样预测 双样预测 MARKOV CHAIN MONTE Carlo 

分 类 号：TB114.3[理学—概率论与数理统计]