【作者】 胡长虹；

【导师】 刘淑芬；

【作者基本信息】 吉林大学 ， 计算机系统结构， 2013， 博士

【摘要】 为了进一步加强计算机软件在互联网络、人工智能、大数据处理、物联网工程、信息传输和图像处理等信息领域的应用能力，加速计算机软件系统的智能化、工程化和产品化的发展进程，突破束缚软件设计跨越式发展的关键技术，提高软件产品的可靠性和安全性。提出了模型驱动架构向可靠性拓扑结构转换方法、不同软件系统拓扑结构的可靠性计算模型及理论、软件系统中功能模块互相重叠的可靠性叠加技术、软件系统基于条件和关联特征结构缺陷的预测方法，使得基于模型驱动架构的软件系统结构可靠性获得大幅度提高；软件系统结构可靠性提升理论有了新的发展和创新。详细分析Weiss、Corcoran、Jelinski、Moranda、Gokhate、Yacoub、毛晓光等人的可靠性模型；Meinhold、Singpurwalla和Ohba等人贝叶斯模型的改进型和非齐次泊松分布模型；cheung等人基于矩阵的状态转移模型；Tarik Hadzic等人的评估模型；徐高潮等人、Khoshgoftaar等人缺陷预测模型等国内外专家学者提出的软件可靠性和软件缺陷预测模型的功能和特点。认真研究和探讨了软件模型驱动架构原理、可靠性增长模型技术、马尔可夫可靠性模型原理与分类和可靠性的缺陷预测方法等内容。经比较确认：所提出的软件可靠性模型理论和方法与这些模型、方法均有较大的区别，属于更高层次的可靠性计算方法与提升技术。提出模型驱动架构向可靠性拓扑结构转换方法。在模型驱动架构向可靠性拓扑结构转换过程中，首次建立不改变对象间的时序关系、不改变对象间的逻辑关系和不改变对象间消息传递的原则。建立了逻辑选择关系的可靠性拓扑结构转换模型、逻辑分支关系的可靠性拓扑结构转换模型和逻辑循环关系的可靠性拓扑结构转换模型。提出了将顺序图中的逻辑选择关系图、逻辑分支关系图和逻辑循环关系图向可靠性拓扑结构转换的方法。试验结果表明：模型驱动架构向可靠性拓扑结构转换方法，能够将系统顺序图准确、合理的转换为可靠性拓扑结构图。提出了可靠性拓扑结构图向可靠性状态图的转换模型和转换方法；给出了各种不同软件系统拓扑结构和可靠性状态转换的可靠性计算理论。依赖系统中模块与模块间的组成方式、复杂度和顺序图，建立了转移概率串行拓扑结构的可靠性模型、转移概率并行拓扑结构的可靠性模型、转移概率串并混合型拓扑结构的可靠性模型、转移概率环状拓扑结构的可靠性模型、多种拓扑结构的可靠性模型；提出了转换概率可靠性拓扑结构与可靠性状态的对应关系、模块执行的转移与模块状态的转移牵动关系、并串混合型结构图与并串混合型的状态图的转换关系；给出了串行拓扑结构、并行拓扑结构、串并混合型拓扑结构和环形拓扑结构的可靠性值计算公式。试验结果表明：不同软件系统拓扑结构的可靠性计算方法均能够计算出软件系统中的可靠度，这些计算结果与系统可靠性的测试结果基本耦合、可信。提出了软件系统中功能模块互相重叠的可靠性叠加技术。经过对软件系统的多分支模块的结构和模块功能的定义与分析，首次建立了软件系统模块分解的原则、功能模块可靠性叠加计算理论和功能模块占用率的可靠性叠加技术；给出了内功能模块与外功能模块可靠性的平均值规划成一个外功能模块的优化可靠性计算方法。并依据结构模型的映射分解法和功能模块的可靠性计算理论，对功能模块占用率可靠性的线性结构进行证明。试验表明：应用可靠性叠加技术原理得到的可靠性计算结果，明显好于未使用叠加技术的可靠性计算结果。提出了软件系统基于条件特征结构缺陷的预测方法。通过对结构缺陷在不同条件下特征和缺陷条件进行分析，提出了缺陷成簇的方法。定义了条件特征和缺陷特征，应用K_means算法对具有相同特征缺陷进行聚类，能够有效和及时的发现缺陷集中出现的特征。建立条件特征与缺陷特征的关系模型和缺陷聚类算法，使具有相同条件的缺陷聚集并成簇。实验验证：基于条件的缺陷聚类算法与K_means算法相结合的结构模型，对缺陷预测的准确度和缺陷的修复效率好于DBSCAN算法，为软件系统缺陷的修复节省了时间和资源，相比传统方法效率提高约30%。提出了软件系统基于关联特征结构缺陷的预测方法。对在同一个缺陷特征簇和不同的缺陷特征簇中的缺陷相似度进行分析，提出了每个簇中均有一个特征重心，代表簇的整体特征趋势的概念；通过对特征重心、缺陷特征和缺陷关联关系属性分析，给出了具有相同特征的缺陷聚集成簇的结论；同时确认缺陷特征不十分相似的缺陷也能聚集成簇。实际数据验证：基于关联特征结构缺陷的预测方法能够进一步为软件修复节省大量的时间和资源，相比传统软件缺陷的修复方法其效率可提高约50%。更多还原

【Abstract】 For improving the application abilities of computer software in internet, artificialintellegence, big data computing, internet of things, data transmission and image processing, andpromoting the developing procedure of intelligence, engineering and products, as well the keytechnology of releasing the tight of software design and realizing great-leap-forwarddevelopment, also including advancing the reliability and safety of software, in this paper wepropose a method of transferring the model driven architecture to topological structure reliability,the theories and models of calculation for different software topological structure reliability, theoverlay reliability technology about overlaps of function modules, defects prediction based ondifferent conditions and related characters. So there is a big rise of architecture softwarereliability on model driven architecture. From the above contributions architecture softwarereliability has a big creation and a dramatic development.We deeply analyze the attributes and functions of software reliability and defects predictionmodels from different researchers including the reliability model of Weiss, Corcoran, Jelinski,Moranda, Gokhate, Yacoub, Xiaoguang Mao, the improved bayesian model andnonhomogeneous Poisson distribution model of Meinhold, Singpurwalla, Ohba, the statetransmit based on matrix proposed by cheung, the measurement model of Tarik Hadzic, thedefects prediction models of Gaochao Xu Khoshgoftaar. Additional we concentrate and discusson model driven architecture, the reliability increase model, Markov reliability model and thedefects prediction based on software reliability. From our compare and methods we confirm thatthe methods and theories proposed in this dissertation have a farther difference with the abovemodels. Our method is a advanced reliability computing method and promotion technology.We propose a method of the transmission from model driven architecture to topologicalstructure reliability. In the procedure of transmission we also present rules of keeping thesequential relationships, keeping logical relationships and keeping the message transmission among the different objects. In this paper we also build three models. Separately they arereliability of topology transformation model based on the logical choice relationship, reliabilityof topology transformation model based on the branching logic and reliability of topologytransformation model based on the logic loop relationship. The method of transferring thepictures of the logical choice relationship, branching logic and the logic loop relationship to thelogic loop relationship is built. The results show that the method of transformation betweenmodel driven architecture and the topological structure reliability can transfer the systemsequence diagram to the picture of topological structure reliability accurately and reasonably.We propose a method of the transformation from the topological structure reliability to thereliability state diagram. Meantime the theories of the different software system topologicalstructures and transformation of the different reliability state diagrams are presented. Dependingon the compositions of different modules, the complexity and sequence diagram, we build thereliability models separately based on transition probability serial topology, transitionprobability of parallel topological structure, transition probability serial and hybrid topology,transition probability of parallel topological structure and kinds of topology structures. thecorresponding relationship between the transformation probability reliability topology structureand reliability state model, the affection relationship between the transfer of models and thetransfer state of models and parallel and the relationship between serial hybrid architecture andserial hybrid states are presented. The formulas of the serial parallel topology, parallel topology,serial and parallel hybrid topology and the loop topology are shown in this paper. Experimentsshow that the different system topology can calculate the reliability reasonably and these resultsand the system reliability test results are basically coupling and credible.In this dissertation a superposition of reliability technology is presented based on thesystem function parts superposition. By multiple branching module structure and modulefunction definition and analyzing on software system, the software system moduledecomposition principle and function module reliability superposition calculation theory and thefunction module usage reliability of superposition technique are proposed. We present using themethod of calculating the mean of inside function module and the outside functional modules tobecome a functional modules. Depending on the method of mapping decomposition on structuremodel and the theory of the reliability calculation about function modules, the linear structure of the reliability function module utilization is proven. The experiment shows that applyingreliability superposition principle can catch the result of the reliability calculation. The effectsare better than the results of without superposition technique.Additional the method of predicting conditions of software system based on thecharacteristics of structure defects is proposed. From analyzing the characters and defectsconditions under the different conditions, the method of defects cluster is proposed. In this partwe also define the condition characters and defects characters. By using K_means to cluster thedefects owing the similar condition characters, the episodes characters can be found. We canbuild condition features and defects relational models to cluster the defects together. Theexperiments show that by combining the cluster theory based on different conditions andK_means, the efficiency and accuracy of this method are better than using DBSCAN. Due tothis method the theory can save the time and resources. Comparing with traditional method a30%rise is presented.Software system prediction method based on correlation characteristics of the architecturedefects is proposed. From analyzing the defects in the same defects cluster and the differentdefects cluster, we propose that in each cluster there is a character center. This center can be onbehalf of the overall characteristics of the cluster. From the analyzation of this center, defectsfeatures and defects relationship, we presented the clusters of the same defects characters.Meanwhile we confirm that the clusters owing un-similar characters also can become theclusters. The experimental data shows that the method based on the related defects architecturecharacters can improve the system fixing time and resources. Comparing with traditionalmethod the efficiency has a50%increase.更多还原