Alarm Ranking Model for Intelligent Management of Metro Systems Based on Statistical Machine Learning Methods

摘要

The Integrated Communication and Supervision (ICS) for Metro Systems responds promptly for detecting defects from thousands of functional devices. However, millions of event messages generated lead to a great amount of errors and nuisance, thus desensitizing the operators' behavior and hindering further maintenance. Therefore, it's desirable to design an Intelligent Alarm Management System(IAMS) for detailed ranking of incidents before being analyzed by human operators. In this paper, a statistical and machine learning based intelligent alarm management system is proposed as a data-driven solution to facilitate the decision making process and forecast disruptions. The alarm ranking model based on the the data from the Singapore metro system consists of data fusion process, ack-delay modeling, and establishment for alarm ranking scores obtained by randomized grid search. The model was constructed using multiple features of both systematical and manual factors from a database composed of 24 million historical incidents and 300 thousand alarms acknowledged. The model has shown high predictive accuracy measured by an adequate validation criterion, as well as good performance in implementation.