研制阶段测试性验证与评价的动态贝叶斯方法  

Testability verification and evaluation method based on dynamic Bayes theory in development stage

作　　者：刘磊[1] 宋家友[1] 姚淼[1] 

机构地区：[1]郑州大学信息工程学院,河南郑州450001

出　　处：《计算机工程与设计》2017年第6期1516-1521,共6页Computer Engineering and Design

摘　　要：针对研制阶段测试性增长实验数据"小子样、多阶段、异总体"的特点导致测试性水平难以验证与评价的问题,提出一种优化的动态贝叶斯方法。引入新Dirichlet分布构造一个故障检测率的动态增长模型;引入D-S区间证据推理理论融合同一阶段的多个专家信息,在此基础上得到置信度更高的先验区间,用非线性优化理论拟合先验信息求解模型中的超参数;利用贝叶斯信息融合理论推断故障检测率的多元联合后验分布,采用Gibbs抽样求解高维后验积分。实例对比分析结果表明,该方法有效地融合了区间型的专家信息,提高了评价结果的置信度,为研制阶段测试性验证与评价的研究提供了一种理论依据和解决方案。To settle the difficulty of verifying and evaluating the testability caused by the growth test data with characteristics of small sample,multi stage,varying population in the development stage,an optimized method based on dynamic Bayes was proposed.The new Dirichlet distribution was used to establish a dynamic growth model of fault detection rate(FDR).The D-S interval evidence reasoning theory was introduced to fuse the expert information at the same stage.The hyper-parameters of the model were calculated based on fitting prior information with high confidence using the nonlinear optimization theory.The Bayes information fusion theory was used to infer the multiple joint posterior distribution of FDR,and Gibbs sampling algorithm was used to calculate the high dimensional posterior integrals.The practical comparison shows that the method proposed can fuse interval expert information effectively and produce the evaluation conclusion with high confidence level,which provides a theory base and solution for the verification and evaluation of the testability in the development stage.

关 键 词：动态贝叶斯 动态增长模型 D-S区间证据推理理论 专家信息融合 非线性优化理论 GIBBS抽样 

分 类 号：TP206.1[自动化与计算机技术—检测技术与自动化装置] TP301.6[自动化与计算机技术—控制科学与工程]