【作者】 王磊；

【导师】 徐东；

【作者基本信息】 哈尔滨工程大学 ， 计算机科学与技术， 2019， 硕士

【摘要】 随着信息技术和信息产业的快速发展,计算机软件系统在众多领域中取得了广泛的应用。近年来计算机软件系统的规模和复杂度不断提高,软件系统的可靠性越来越难以度量。软件可靠性度量技术经过多年发展,众多专家学者提出了几百种适合不同应用场景的软件可靠性度量模型来度量软件系统的可靠性。论文通过对软件可靠性增长模型的理论研究和实践,针对基于GRU(Gated Recurrent Unit)神经网络的软件可靠性增长模型存在的问题,提出一种深度CG-EGU(Efficient Gated Unit with Control Gate)神经网络,并且利用深度CG-EGU神经网络,在软件失效数据集上建立软件可靠性模型。论文的研究工作分为以下三个部分:(1)针对GRU单元更新门学习能力不足导致GRU单元整体学习效率低的问题,提出了一种高效门控单元EGU(Efficient Gated Unit)。为了使GRU单元更新门学习时更有效,EGU单元在GRU单元的更新门更新时引入遗忘门,增加了更新门更新的效率,降低了训练开销,提高模型性能。(2)针对深度GRU网络多个隐含层之间信息处理效率不高的问题,提出一种基于深度EGU网络的深度CG-EGU神经网络。深度CG-EGU神经网络在深度EGU网络的基础上,通过在不同隐含层之间添加控制门的方式,增强隐含层之间的信息处理能力,进而提高信息处理的效率。(3)通过实验对本文提出的EGU单元和深度CG-EGU神经网络的性能分别进行对比验证。在公认的软件缺陷数据集中,分别建立基于深度BP神经网络、深度GRU神经网络、深度EGU神经网络、深度CG-EGU神经网络的软件可靠性增长模型,通过比较模型的召回率,验证在隐含层个数相同时,EGU单元相比于GRU单元而言性能是否提高。验证在隐含层个数不同时,深度CG-EGU网络相比于深度EGU网络而言性能是否提高。更多还原

【Abstract】 With the rapid development of information technology and information industry,computer software systems have been widely used in many fields.In recent years,with the continuous improvement of the scale and complexity of computer software systems,the reliability of software systems is increasingly difficult to secure and measure.With the development of software reliability measurement technology for many years,many excellent software reliability and growth models have emerged to measure software reliability.Based on the theoretical research and practice of software reliability growth model,this paper proposes a deep CG-EGU(Efficient Gated Unit with Control Gate)neural network based on the deep GRU(Gated Recurrent Unit)neural network aiming at the problem of software reliability growth model based on GRU neural network.The paper uses the deep CG-EGU neural network to establish a software reliability growth model on the software failure data set to measure software reliability.The research of this paper is divided into the following three parts:(1)Aiming at the problem that the GRU unit has low learning efficiency due to insufficient GRU unit update learning ability,a highly efficient EGU(Efficient Gated Unit)is proposed.In order to make the GRU unit update gates more efficient,the EGU unit introduces a forgotten gate when the GRU unit updates the gate update,which increases the efficiency of updating the gate update,improves the training efficiency of the GRU unit,and reduces the training overhead.(2)For the problem that the information processing efficiency between multiple hidden layers in the deep GRU network is not high,this paper proposes a deep CG-EGU neural network based on deep EGU network.On the basis of the deep EGU network,the deep CG-EGU neural network enhances the information processing capability between the hidden layers by adding control gates between different hidden layers to improve the efficiency of feature extraction and further improve the performance of the model.(3)The performance of the EGU unit and the deep CG-EGU neural network proposed in this paper were compared and verified by experiments.Software reliability growth models based on deep BP neural network,deep GRU neural network,deep EGU neural network anddeep CG-EGU neural network are established in the recognized software defect dataset.Compare and analyze the prediction results of the software reliability growth model from the same and different perspectives of the hidden layer.Verify whether the performance of the EGU unit is higher than that of the GRU unit when the number of hidden layers is the same.Verify whether the performance of the deep CG-EGU network is improved compared to the deep EGU network when the number of hidden layers is different.更多还原