Outlier handling

Historical data collected from meters, sensors and other building automation devices can skew calculations if records are missing or contain invalid values. Data may contain unwanted or inaccurate values caused by a sudden electrical current surge, meter reset or device failure. Operations performed on incomplete data sets and on records that contain junk, outlier, or noise values inevitably produce inaccurate results. Beginning with a recent version of Niagara, you can select which history records to exclude from the data set based on record Status..

For example, the following table contains invalid data.

Timestamp Value Status Description
15/12/21 2:00 99999099 {ok} This is an extremely high value, which is invalid.
15/12/21 3:00 10 {ok} This is a valid value.
15/12/21 4:00 20 {ok} This is a valid value.
15/12/21 5:00 NaN {ok} This value is invalid because it is not a number.

Not a Number, expressed as NaN, is an actual numeric value similar to positive infinity expressed as +inf or negative infinity expressed as -inf.

15/12/21 6:00 20 {fault} This value is valid, but the status of the device indicates a problem.
15/12/21 7:00 20000 {ok} This invalid high value was caused by a sudden meter reset.
15/12/21 8:00 20010 {ok} This is another invalid high value caused by a sudden meter reset.

Based on device status, you can filter out the records that contain invalid data. This creates a data set with missing records. You then use a missing data strategy to interpolate the missing data.