基于贝叶斯网的分布式软件行为运行时可信性分析

Analyzing Running-time Behavioral Credibility for Distributed Software Based on Bayesian Network

在开放、动态和复杂的网络环境中,监测与分析软件行为可信对现代分布式软件是至关重要的.针对分布式软件运行时的外在表现特征,系统地收集相关数据,根据具体交互场景建立贝叶斯网模型.利用此模型,在上下文环境中通过监测相关的数据来对软件行为运行时可信性进行分析.建网过程中,文中提出了使用"3σ原则"来离散化连续型随机变量,其在判断样本标注异常及先验参数确定等方面具有独特优势,操作方便又符合实际情况,且提高了参数学习效率;同时,文中提出了分层方法构造先验贝叶斯网思想,通过计算节点间的相关系数来逐步修正贝叶斯网结构,降低了建网的复杂性和误差.通过仿真实验,证实了本文所提出的方法在软件行为可信性分析方面较其他方法有着独特的优势.

In an open,dynamic and complex internet environment,monitoring and analyzing behavioral credibility is essential for the modern distributed software.It is a good idea to build a Bayesian network with specific interaction scenarios through collecting relevant data systematically,which can analyze software behavior′s credibility in a specific context for the running-time performance with some external features of distributed software.In network-building process,this paper proposes a so-called ＂3σ principle＂ to make the continuous random variable be discrete,which has two unique advantages in marking sample and determining the priori parameters of Bayesian learning.The ＂3σ principle＂ accords to actual situation of running-time software and operates easily,which can improve efficiency in the parameter learning.At the same time,this paper proposes a method to construct a Bayesian network hierarchically by calculating the correlation coefficient between two nodes in order to modify the network structure,which can reduce the complexity and shrink the error during building network.The experiments illuminate that the method proposed in this paper has unique advantage in credible analysis of software behavior.