Automatic train supervision (ATS) systems are designed to improve the reliability of train services. An ATS system coordinates the trains and other systems in a metro and records alarms if faults occur. In this work, we propose a context-aware anomaly diagnosis tool to analyze the underlying causes of alarms for ATS system. Using 61-day data collected from an operational ATS system, we apply our diagnosis tool to conduct systematic analysis of the alarms and identify interesting correlations among different assets and events. Our analysis shows that the alarms can be correlated with certain system events if they are in the same operations or the assets associated with them belong to the same or linked systems. These results can improve the efficiency of anomaly diagnosis and maintenance for metro system.