A Dynamic Screening System for Early Detection of Multiple Interconnected Events

Sequential monitoring of temporal processes is essential in various fields for the early detection of critical events, such as airplane failures or disease occurrences. This task typically involves sequential decision-making across many related processes. Conventional statistical process control charts are generally designed for detecting an event of interest in a single process and thus inadequate for this task. Recently, several versions of the dynamic screening system (DySS) have been developed to monitor a population of processes. However, these DySS methods focus solely on the early detection of a single event (e.g., the occurrence of a particular disease) and cannot handle scenarios where multiple distinct events are involved. In practice, detecting multiple events is important and common. For example, we may be concerned with several medical conditions in patients or different types of failures in airplanes. This problem is significantly more challenging than single-event detection, as different events are often interconnected and can occur at different times. To address this complexity, we propose the concept of conditional risks for multiple events, using single-index multinomial logistic regression modeling. Based on this, we develop a new DySS method for the early detection of multiple events by sequential monitoring of the related conditional risks. Numerical studies demonstrate that this method provides an effective analytical tool for the early detection of multiple interconnected events.

时序过程（temporal processes）的序贯监测（sequential monitoring）在诸多领域中对于关键事件的早期预警至关重要，例如飞机故障或疾病发作。此类任务通常需在多组关联的时序过程中实施序贯决策。传统统计过程控制图（Statistical Process Control Charts）通常仅针对单一时序过程中的目标事件设计检测方案，因此无法适用于此类多过程监测任务。近年来，已有研究者开发出多款动态筛选系统（Dynamic Screening System，DySS），用于对一批时序过程群体开展监测。然而，现有DySS方法仅聚焦于单一事件的早期检测（例如某一类特定疾病的发作），无法处理涉及多种不同事件的监测场景。实际应用场景中，多事件检测不仅必要且极为常见。例如，临床中需同时监测患者的多种病症，航空领域则需侦测飞机的各类故障。相较于单事件检测，多事件监测问题的挑战性显著提升，因为不同事件往往相互关联且可能在不同时刻发生。为解决此类复杂问题，本文基于单指标多项Logistic回归（single-index multinomial logistic regression）建模，提出了多事件条件风险的概念。基于此概念，本文通过对相关联的条件风险实施序贯监测，提出了一种全新的DySS多事件早期检测方法。数值实验结果表明，所提方法可为多关联事件的早期检测提供高效的分析工具。