【作者】 孙莉；

【导师】 徐宝文；

【作者基本信息】 东南大学 ， 计算机软件与理论， 2006， 硕士

【摘要】 70年代初期,在Jeliski-Moranda提出第一个软件可靠性模型后,有关软件可靠性的研究就一直是软件工程领域的热点问题。经过近几十年的研究和发展,软件可靠性和软件可靠性工程的理论逐渐成熟,其中最重要的研究方向之一就是组件软件可靠性以及组件软件可靠性的优化问题。组件软件可靠性一般无法准确测量,所以如何准确估计或预测组件软件可靠性是组件软件可靠性研究中的重要问题。由于用来估计组件软件可靠性的可靠性数据来源各异,使得组件软件可靠性估计结果存在着不可靠性,组件软件可靠性估计必定会产生估计方差。我们在实际环境中分析组件软件可靠性的时候,应该考虑到可靠性估计方差因素,这样才能够保证可靠性估计结果的正确性和完备性,开发和应用真正有意义的组件软件可靠性估计模型和组件软件可靠性优化模型。组件软件可靠性优化的主要任务就是在提高可靠性的同时,还需要考虑测试代价的以及可靠性估计方差的最小化。然而,传统的RPP组件软件可靠性优化策略并没有考虑到测试代价对优化结果的影响。可靠性估计方差最小化问题虽然已引起人们的注意,但也一直缺乏比较简单和准确的解决方法。本文的创新之处首先在于提出了一种改进的RPP组件软件可靠性优化策略:RPP-c组件软件可靠性优化策略。RPP-c优化策略在降低RPPI值高的组件软件可靠性估计方差的同时,增加考虑了与该组件软件相关的额外的测试代价因素。其最终解应该由组件软件的RPPI值和额外测试代价共同决定,即在较低的成本水平下选取对软件系统整体可靠性估计贡献较大的组件软件进行额外测试,以便提高整个软件系统可靠性水平。本文简单给出RPP-c组件软件可靠性优化策略的算法实现思想。本文的另一个创新之处是提出采用交叉熵方法解决最小化组件软件可靠性估计方差的方案。为了减少可靠性估计方差给组件软件可靠性研究带来的各种影响,最小化可靠性估计方差的任务迫在眉睫。交叉熵方法为处理可靠性优化问题提供了一种快速、实用并且行之有效的方法。我们利用交叉熵的概念和特性,通过拟合方法和重要抽样方法,提出用交叉熵方法最小化可靠性估计方差的改进方案。本文具体给出交叉熵最小化可靠性估计方差的算法实现步骤。更多还原

【Abstract】 When Jeliski-Moranda presented the first model of software reliability in the early days of 70’s,the research about software reliability is the highlight of software engineering all the while。By recent investigation,software reliability and software reliability engineering became gradually maturity,and the most important aspect is the research about component software。Because we can not exactly measure reliability of component software,how to give the precision reliability estimation of component software is the main problem。The data source used to estimate vary,so the uncertainty of component software reliability must exist,and there are estimation variance in component software reliability。Thus,it is great need to consider the factor of estimation variance to develop truly useful reliability estimation models and reliability optimization of component software。There have two innovations in the thesis: RPP-c strategy and the use of Cross-Entropy to solve the problem of minimizing estimation variance of component software.The traditional RPP strategy of component software reliability optimization does not consider the cost of the additional testing required to decrease component-estimate variance。So we put forward a modified RPP strategy:RPP-c strategy。We should not only think over how to decrease the estimation variance of component software with higher RPPI,but also need to calculate the interrelated supererogatory test cost。The RPPI value and the test cost determine the ultimately solution together,namely we ought to choose these component software which can contribute much to the whole system reliability estimation under much lower cost level。In order to reduce the influence of reliability estimation variance,we must consider the problem about minimizing estimation variance。It is a celerity、utility and effective method to use Cross-Entropy to deal with reliability optimization。With theory of Important Sample(IS) and Cross-Entropy(CE),we bring forward the project to minimize estimation variance。At last,we give the CE algorithm and experiment result of component software optimization。更多还原