一种面向CPS的自适应统计模型检测方法

发布时间：2018-06-18 16:14

本文选题：信息-物理融合系统 + 统计模型检测　；参考：《软件学报》2017年05期

【摘要】：随着计算机与物理环境的交互日益密切,信息-物理融合系统(cyber-physical system,简称CPS)在健康医疗、航空电子、智能建筑等领域具有广泛的应用前景,CPS的正确性、可靠性分析已引起人们的广泛关注.统计模型检测(statistical model checking,简称SMC)技术能够对CPS进行有效验证,并为系统的性能提供定量评估.然而,随着系统规模的日益扩大,如何提高统计模型检测技术验证CPS的效率,是目前所面临的主要困难之一.针对此问题,首先对现有SMC技术进行实验分析,总结各种SMC技术的受限适用范围和性能缺陷,并针对贝叶斯区间估计算法(Bayesian interval estimate,简称BIE)在实际概率接近0.5时需要大量路径才能完成验证的缺陷,提出一种基于抽象和学习的统计模型检测方法 AL-SMC.该方法采用主成分分析、前缀树约减等技术对仿真路径进行学习和抽象,以减少样本空间;然后,提出了一个面向CPS的自适应SMC算法框架,可根据不同的概率区间自动选择AL-SMC算法或者BIE算法,有效应对不同情况下的验证问题;最后,结合经典案例进行实验分析,实验结果表明,自适应SMC算法框架能够在一定误差范围内有效提高CPS统计模型检测的效率,为CPS的分析验证提供了一种有效的途径.
[Abstract]:With the increasingly close interaction between computer and physical environment, cyber-physical system (CPSfor short) has a wide application prospect in health care, avionics, intelligent building and so on. Reliability analysis has attracted wide attention. Statistical model checking (SMC) technology can effectively verify the performance of statistical model and provide quantitative evaluation for the performance of the system. However, with the increasing scale of the system, how to improve the efficiency of statistical model detection technology to verify CPS is one of the main difficulties. In order to solve this problem, the existing SMC technology is analyzed experimentally, and the limited application scope and performance defects of various SMC technologies are summarized. Aiming at the defects of Bayesian interval estimation (BIEs) algorithm, which requires a large number of paths to complete verification when the actual probability is close to 0.5, an abstract and learning-based statistical model detection method, AL-SMC-based, is proposed. In this method, principal component analysis and prefix tree reduction are used to study and abstract the simulation path to reduce the sample space. Then, an adaptive SMC algorithm framework for CPS is proposed. AL-SMC algorithm or BIE algorithm can be automatically selected according to different probabilistic intervals, which can effectively deal with the verification problem under different conditions. Finally, the experimental results show that, The adaptive SMC algorithm framework can effectively improve the efficiency of CPS statistical model detection within a certain error range, which provides an effective way for the analysis and verification of CPS.
【作者单位】：教育部可信软件国际合作联合实验室(华东师范大学);可信软件国际联合研究中心(华东师范大学);上海市高可信重点实验室(华东师范大学);
【基金】：国家自然科学基金(61472140,61170084) 上海市自然科学基金(14ZR1412500)~~
【分类号】：TP311

【参考文献】