基于随机自动机状态估计的故障预测  

State estimation based fault prediction for stochastic automatons

作　　者：常明[1] 董炜[1,2] 吉吟东[1,2] 张曾科[1] 

机构地区：[1]清华大学自动化系,北京100084 [2]清华大学信息科学与技术国家实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2013年第11期1623-1628,1636,共7页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(61104019;61004070);国家科技支撑计划项目(2009BAG12A08);清华大学自主科研计划资助项目(2011Z06121;2012Z06115)

摘　　要：逻辑自动机下的可预测性分析趋于保守,通常在实际系统中应用受限。该文研究基于随机自动机的故障预测问题。对于每一个正常的系统状态,应用吸收概率理论计算其转移到故障状态的概率和平均时间。根据系统事件的可观测性构造诊断器,确定系统可能处于的状态集合。基于观测序列,确定系统状态分布,通过概率加权计算系统转移到故障状态的概率和平均时间。应用HVAC(heating,ventilation and air conditioning)系统的仿真实例验证算法的有效性。结果表明:该方法能够预测不同观测序列下系统发生故障的概率和平均时间。此外,对于逻辑不可预测系统,该方法依然适用。Logical predictable conclusions tend to be conservative, which always imposes some restrictions in practical applications. This analysis investigates fault prediction in stochastic automatons is investigated. The probability that a system in each normal state will transition to a fault state for the first time is computed using absorbing probability theory, along with the average transfer time. A diagnoser is constructed according to the observability of system events to identify the set of possible states. Observation sequences are used to compute state distributions with the probability and the average transfer time for system to transition to a fault state obtained through probabilily weighting. A heating, ventilation and air conditioning （HVAC） system is analyzed to demonstrate the feasibility of this method. The system can predict the occurrence probability and the occurrence time of fault events for observation sequences. This method is also applicable to non-logical predictable cases.

关 键 词：离散事件系统 故障预测 随机自动机 诊断器 吸收概率理论 

分 类 号：TP301.1[自动化与计算机技术—计算机系统结构]