基于错误发现率的高维数据流在线监控方法

Online Monitoring Method of High-Dimensional Data Streams Based on False Discovery Rate

关于多数据流的监控,大多假设数据流之间是独立的。从统计过程控制的角度,给出了在线监控高维数据流的一般框架。鉴于数据的分布可能存在多样性,本文采用对称数据聚合方法建立了稳健的监控统计量,利用统计量的渐进对称性选取数据驱动的阈值,基于错误发现率对相关的非正态数据流进行在线监控。以AR (1)模型刻画数据流间的相关性,通过蒙特卡洛模拟,研究了所提出方法的错误发现率和功效水平。数值模拟结果表明所提出的方法具有较理想的性能。

Regarding the monitoring of multiple data streams, it is mostly assumed that the data streams are independent. A general framework for online monitoring of high-dimensional data streams is provided from the perspective of statistical process control. Given the potential diversity in data distribution, this paper adopts a symmetric data aggregation method to establish a robust monitoring statistic. The asymptotic symmetry of the statistic is used to select data-driven thresholds, and the relevant non-normal data streams are monitored online based on the false discovery rate. The AR (1) model was used to characterize the correlation between data streams, and the false discovery rate and power level of the proposed method were studied through Monte Carlo. The numerical simulation results indicate that the proposed method has ideal performance.