Anomaly Detection | HydroNode

01Three kinds of checks

What a reading is compared with

Your targets

A min and max you set per measurement. Active from the first reading.

Guidelines

Published limits such as the WHO and UBA values for CO₂, particulate matter or UV. Active from the first reading.

Learned band

The range this sensor usually shows at this time of day. Needs some history first.

An unusual value is not automatically a problem. Only a broken target or guideline can make an incident critical. A value that is merely unusual reaches Warning at most and never asks you to act.

02Learning

How HydroNode learns what is normal

For each measurement HydroNode builds a model from the last 28 days of hourly averages. Recent days count more than old ones. It learns a typical level and, once there is enough data, a daily pattern, so a warm afternoon is not flagged just because the night was cool. The model is rebuilt every 15 minutes.

Stage	Needs	What changes
First reading	Nothing	Targets and guidelines are checked.
Learned band	20 readings over at least 6 hours	Statistical checks start.
Daily pattern	3 days with good coverage	Time of day is taken into account.
Full pattern	7 days	The daily pattern is fully trusted.

Single outliers do not shift what counts as normal. A new level that stays stable for 6 hours is accepted, for example after you move the sensor. A new level that breaks a target or guideline is never accepted: air that is always bad stays bad.

03Incident types

What HydroNode can find

Type	When it triggers
Abrupt change	A sudden jump far outside the learned band. It only counts if the next reading confirms it, so a single glitch is ignored.
Persistent deviation	The value stays clearly outside the band for at least 15 minutes.
Gradual drift	A slow, steady shift in one direction over several hours. This describes the data; it does not claim the sensor needs calibration.
Target breach	The value is outside the target you set.
Reference breach	A published guideline is exceeded, for example CO₂ above a ventilation value.
Recovery failure	An expected return does not happen. Example: humidity is still above 65 % an hour after a shower.
Data quality	Impossible values, or a value that normally moves but has not changed in 6 hours and 30 readings.

Some patterns are recognised but are not anomalies: likely ventilation, watering, a shower, battery charging, a counter reset and the normal daily cycle. They give context and do not alert you.

Measurements without statistical checks

Weather and sky values change with the weather, not with your setup: UV, solar radiation, light, wind, rain and air pressure. They are not judged against a learned band, but guidelines such as the WHO UV index still apply. Pollutants like CO₂ and particulate matter are only flagged when they rise, never when the air gets cleaner.

04Incidents

One incident per problem

A problem that keeps going stays one incident. New readings update it instead of opening new ones. Open incidents are listed in the anomaly panel on the right of the sensor page.

1
Open
The problem was confirmed.
2
Ongoing
It is still there. Severity can go up; it never silently goes down.
3
Recovering
Values are back in range.
4
Resolved
After at least two good readings and 10 minutes. Missing data never resolves an incident.

Click an incident to see the trigger value, the rule, the readings around it and what to do.

05Tuning

What you can change

Only the owner can change these settings. People you shared the sensor with see the same results.

Sensitivity

Set on the Anomaly detection tile of the sensor page, or in Sensor Settings → Analysis. It widens or narrows the learned band. The tile also shows how many incidents the last 7 days produced and may suggest a calmer level.

Level	Band	Good for
Very relaxed	2.0×	Only extreme outliers.
Relaxed	1.5×	Clear deviations only.
Standard	1.25×	Default. Balanced for rooms and gardens.
Strict	1.0×	Flags smaller deviations.
Very strict	0.8×	Flags every noticeable change.

Sensitivity only affects the learned band. Your targets and published guidelines always stay exactly as they are.

Target bands

Open Target bands & alerts on the sensor page and set a min and max per measurement. The learned band is shown next to it as a reference, and where a guideline exists HydroNode offers it as a one-click suggestion.

Environment and calibration

In Sensor Settings → Analysis, choose Indoor, Outdoor or Greenhouse (treated like indoor). This changes what counts as normal. You can also set the time zone for daily patterns, a battery type, and a calibration offset per measurement if your reference meter shows a different value. Try it on past data shows how many alerts your ranges would have raised, without changing anything.

Rate incidents

Each incident can be rated Normal, Issue or Fault. When the owner rates an incident as Issue or Fault, that time is left out of learning, so the problem never becomes the new normal. Normal lets HydroNode learn from it. Ratings by people you shared the sensor with stay personal and do not change the model.