基于程序状态分析的核函数软件失效预测关键技术研究

Software reliability modeling is one of the most efficient tools for improving the software reliability. Its significant research lies in building a self-adaptive prediction model of the software failure time, with its dynamics and complexity taken into consideration. Programming behavior includes tons of information, and an analysis on the state pattern of the software fault behavior during the program running may reflect the characteristics of software failure. The research, based on fault injection, is to design a group of clustering experiments, so as to collect a set of data about programming behavior and software failure time. Next, the research is to adopt kernel regression theory, and on the basis of quantitative analysis, to explore the ways of establishing interrelations between programming behavior pattern and software failure, particularly the failure time. In addition, we will proceed the research of how to choosing or constructing the kernel functions, the regression estimation algorithms, and the automatic assignment algorithm for different parameters of the model. The research is also to explore the ways of improving the prediction capability of software failure time and adaptive ability of the model under the condition of inadequate failure data. And it is expected to establish with a completely theoretical system of software reliability modeling, and help enrich our knowledge about the complexity of software failure behavior.

目前提升软件可信性水平最有效的工具之一为软件可靠性建模。程序运行状态中包含了许多对软件失效行为预测有益的动态信息，本项目以程序运行状态序列数据、软件失效行为数据为对象，研究复杂结构数据核学习的新模型和新算法，以核函数学习理论为主要工具开发软件失效行为预测的有效方法。主要研究为：(1)基于故障注入的程序运行状态和软件失效行为数据收集；(2)程序运行状态序列数据非线性子空间聚类算法设计；(3)基于程序状态序列特征选择的多核软件失效行为预测模型。项目探索从定量分析的角度建立程序行为状态与软件失效行为特别是软件失效时间之间的联系，通过对核函数自适应构建以及核参数动态赋值算法的研究，切实提高软件失效数据不充分情况下的软件失效行为预测的精确性及适用能力，提升对软件失效行为复杂机理的认知水平。

本项目以程序运行状态序列数据、软件失效行为数据为对象，研究了复杂结构数据核学习的新模型和新算法，以核函数学习理论为主要工具开发软件失效行为预测的有效方法。主要研究内容包括：基于故障注入的程序运行状态和软件失效行为数据收集；程序运行状态序列数据非线性子空间聚类算法设计；基于程序状态序列特征选择的多核软件失效行为预测模型。项目探索了从定量分析的角度建立程序行为状态与软件失效行为特别是软件失效时间之间的联系，通过对核函数自适应构建以及核参数动态赋值算法的研究，切实提高了软件失效数据不充分情况下的软件失效行为预测的精确性及适用能力，提升了对软件失效行为复杂机理的认知水平。通过本项目的研究，在TC、TCAD、TNNLS、JCST、DATE等高水平期刊或者会议发表或录用论文34篇，其中SCI、EI检索34篇，包含IEEE Trans系列论文14篇，CCF A、B类论文15篇；授权发明专利2项，申请发明专利4项，获得软件著作权3项；项目组成员晋升教授2人、晋升副教授2人、获得浙江省杰出青年基金资助1人、入选浙江省万人计划1人、入选浙江省高校领军人才培养计划2人、入选南太湖特支计划3人，目前在读研究生30余人。该项目的研究丰富了软件可靠性建模的理论和方法，对揭示软件失效过程的规律有重要意义。