小子样变总体下的Bayes测试性验证方法  

Testability verification method based on Bayes theory under small sample and varying population circumstance

作　　者：汤巍[1] 景博[1] 黄以锋[1] 

机构地区：[1]空军工程大学航空航天工程学院,陕西西安710038

出　　处：《系统工程与电子技术》2014年第12期2566-2570,共5页Systems Engineering and Electronics

基　　金：国家自然科学基金(51201182);航空科学基金(20101996012)资助课题

摘　　要：现有测试性验证试验方案一般都需要较大的故障样本量,利用装备研制阶段的测试性数据则可能有效解决这一问题。但研制阶段的测试性数据和现场试验数据一样属于"小子样"且具有"变总体"特点,为此提出一种基于Bayes理论的测试性验证方法。首先根据装备研制各阶段积累的测试性试验信息构建装备测试性参数故障检测/隔离率(fault detection rate/fault isolation rate,FDR/FIR)的动态增长模型,用以揭示装备测试性水平的动态变化规律,并对其测试性指标进行预测。然后根据最大熵原理计算系统FDR/FIR的先验分布。最后结合少量现场试验数据,根据Bayes最大验后风险准则制定装备测试性验证方案,对装备测试性指标进行验证。通过实例对比分析表明,该方法有效融合了研制阶段数据与现场试验数据,能在小样本的情况下提高验证结论的置信度,降低评估风险。The fault sample size which is required in existing testability demonstration test schemes may be reduced by using data in the development phase.However,the growth test data are“small samples”with“varying population”.A new testability verification method based on the Bayes theory is proposed.Firstly,the proposed method establishes a dynamic growth model of the test parameters based on multiple phases’samples,which is used to describe the changing rule of equipment’s testability and predict the fault detection rate （FDR）and the fault isolation rate （FIR）.Then,the prior distribution of the system’s FDR/FIR is calculated based on the maximum entropy principle.Finally,a new testability determination scheme is defined to verify FDR/FIR,ac-cording to Bayes maximum posterior risk rules with a small size of field trial data.The practical comparison shows that this method,in which the growth test data and field trial data can be fused effectively,can reach an evaluation conclusion with a high confidence level under small sample circumstance,and reduce the risk of evaluation.

关 键 词：测试性验证 Bayes理论 动态增长模型 小子样 变总体 

分 类 号：TP302.8[自动化与计算机技术—计算机系统结构]