研制阶段测试性验证与评价的动态贝叶斯方法

发布时间：2018-05-31 01:01

  本文选题：动态贝叶斯 + 动态增长模型　； 参考：《计算机工程与设计》2017年06期

【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。
[Abstract]:In order to solve the problem that the testability level is difficult to verify and evaluate due to the characteristics of "small sample, multi-stage, different population" in the experimental data of testability growth in the development phase, an optimized dynamic Bayesian method is proposed. The new Dirichlet distribution is introduced to construct a dynamic growth model of fault detection rate, and D-S interval evidential reasoning theory is introduced to fuse multiple expert information in the same stage, and on this basis, a priori interval with higher confidence is obtained. The nonlinear optimization theory is used to fit the transcendental information and the Bayesian information fusion theory is used to infer the multivariate joint posterior distribution of the fault detection rate and the Gibbs sampling is used to solve the high dimensional posterior integrals. The results of comparison and analysis show that the method effectively integrates the interval type of expert information and improves the confidence of the evaluation results. It provides a theoretical basis and a solution for the research of testability verification and evaluation in the development phase.
【作者单位】： 郑州大学信息工程学院;
【分类号】：O212.8
，

本文编号：1957607