Anomaly Detection: Six Methods, One Queue

A modern building generates tens of thousands of signals an hour. A small fraction of those are anomalous. A small fraction of those are actually worth your attention. Anomaly detection is the discipline of producing that small fraction reliably.

FrostLogic Explore detects across six methods: Sensor drift — slow, sustained deviation from the expected baseline. Stuck values — sensor returns a constant where variance is expected. Energy spike — statistically improbable consumption jump. Forecast deviation — reality has diverged from the forecast beyond confidence band. Sensor offline — reporting cadence stopped or degraded. Cross-signal incoherence — related signals physically contradict each other.

Every anomaly is severity-scored, classified, and dropped into a single prioritised queue — the inverse of the alarm storm a traditional BMS produces.

This is the detection layer inside energy monitoring: FrostLogic Explore's real-time reads feed straight into these six methods, and the ranked output lands in the same queue operators already work from.

The same six methods are also what drive predictive maintenance: forecasting a developing fault days or weeks before it trips a threshold, instead of just flagging it after the fact.