The distributed nature of the power grid, and of its regulatory and supervisory system, makes user adoption of automated technologies even harder than usual. Perceived utility and perceived ease of use, the two principal drivers of user adoption, are harder to achieve in a diverse environment in which users will have different scope of responsibility and skill levels. The difficulties in anomaly detection will set a high floor to the false alarm rate. A large number of false alarms, that need to be responded to by a distributed heterogenous team of supervisors, can easily trigger in supervisors a "cry-wolf" effect, rendering the detection system useless. To achieve the objective of helping power grid operators build and maintain situational awareness, without succumbing to user rejection, Barnstorm Research proposes to develop RAMI, an anomaly detection triage engine. To avoid triggering the cry-wolf effect, RAMI will sort the predicted adverse events into detection classes according to their impact on supervisor actions. The detection classifier will analyze the predicted failure mode and the corresponding recovery actions along multiple complexity metrics (duration, geographical scope, severity). The detection classifier will use class-specific action thresholds, tailoring the false alarm rate to class-specific ability to mitigate its effects.
