Alarm processing with model-based diagnosis of event discrete systems

Reliable and informative alarm processing is important for improving the situational awareness of operators of electricity networks and other complex systems. Earlier approaches to alarm processing have been predominantly syntactic, based on text-level filtering of alarm sequences or shallow models of the monitored system. We argue that a deep understanding of the current state of the system being monitored is a prerequisite for more advanced forms of alarm processing. We use a model-based approach to infer the (unobservable) events behind alarms and to determine causal connections between events and alarms. Based on this information, we propose implementations of several forms of alarm processing functionalities. We demonstrate and evaluate the resulting framework with data from an Australian transmission network operator.