基于控制过度遗漏发现概率的高维数据流异常诊断

High-dimensional Fault Diagnosis by Controlling Missed Discovery Excessive Probability

随着科技的创新和社会的进步,数据采集技术得到显著提升,高维数据流(High-dimensional Data Stream,HDS)在医学、机械、工业工程等领域开始广泛出现。除了HDS的在线监控之外,精确而高效的故障诊断也变的越来越重要。在本文中,我们将HDS的故障诊断问题阐述为一个新颖的多重检验问题,并通过控制过度遗漏发现概率(Missed Discovery Excessive Probability,MDX)来对HDS进行异常诊断,克服了传统诊断方法的限制,并能够显著的提高异常诊断的稳健性和精确度。我们给出了MDX的Monte-Carlo近似计算方法,并在此基础上提出了Oracle和DataDriven诊断程序。我们通过模拟研究和一个实例分析来阐明所提方法的优越特性。

With the development of science and technology,data collection technology has been improved significantly,and high-dimensional data streams(HDS)has been widely seen in many domains such as medication science,mechanical engineering and industrial engineering.In addition to HDS online monitoring,accurate and efficient fault diagnosis has also become more and more important.In this paper,we describe the fault diagnosis of HDS as a novel problem of multiple testing,and propose a fault diagnosis procedure by controlling missed discovery excessive probability(MDX),which overcomes the limitations of traditional diagnosis methods and can significantly improve the robustness and accuracy of fault diagnosis.We provide the Monte-Carlo approximate calculation method of MDX,and then propose Oracle and Data-Driven diagnostic programs based on this method.We demonstrate the superior characteristics of the proposed diagnostic method in practical applications through simulation studies and an actual data analysis.