How does pattern learning work and what does the system learn from my data?

Pattern alarms rely on a trained consumption model — a statistical profile of your meter's normal behavior. Before a pattern alarm can work, you must train this model. Here is how the process works:

Training process

Go to Analytics → Alarms → select your meter → click "Learn Pattern"
The system pulls 8 weeks of historical readings for that meter
It requires at least 100 data points to build a reliable model — if your meter has less, you will see an error

What the model learns

Hourly pattern (24 values): Your average consumption for each hour of the day. For example, it might learn that electricity is lowest at 3 AM and peaks at 10 AM. This captures your building's daily rhythm.
Daily pattern (7 values): Your average consumption for each weekday. Monday through Friday may show high usage while Saturday and Sunday are near zero — common for office buildings.
Seasonal factors (4 values): Multipliers for winter (Dec–Feb), spring (Mar–May), summer (Jun–Aug), and fall (Sep–Nov). A heating meter might have winter = 1.3 (30% above average) and summer = 0.6 (40% below). These adjust the baseline so the alarm does not trigger simply because winter arrived.
Statistical bounds: The model calculates mean usage and standard deviation. The upper bound is set at mean + 2.5 standard deviations, and the lower bound at max(0, mean - 2.5 standard deviations). Only about 0.6% of normal readings should fall outside these bounds.

How the alarm uses the model

When the evaluator checks a pattern alarm, it takes the current hour's pattern value, applies the daily and seasonal adjustments, and compares the actual reading. If the reading deviates more than the configured percentage (default 25%) OR falls outside the statistical bounds, the alarm triggers.

Retraining: Models do not update automatically. If your building's usage changes significantly (e.g., new tenants, equipment changes), retrain the model by clicking "Learn Pattern" again.

Where to manage: Analytics → Alarms → select a meter → Pattern Learning section.