[Feature]: Dealing with faulty (non-circular) averaging in SCADA

NREL / flasc

A rich floris-driven suite for SCADA analysis

https://nrel.github.io/flasc/

BSD 3-Clause "New" or "Revised" License

28 stars 18 forks source link

[Feature]: Dealing with faulty (non-circular) averaging in SCADA #90

Open Bartdoekemeijer opened 1 year ago

Bartdoekemeijer commented 1 year ago

Description

FLASC currently assumes that the wind direction is the underlying SCADA has been correctly time-averaged, i.e., dealing with 360 deg wrapping for angular variables like wind direction and nacelle heading. However, this is often not the case when working with 10-minute averaged data from an operational asset. It would be great if FLASC could find these obvious outliers and mark them as faulty.

Related URLs

No response

paulf81 commented 1 year ago

That's a nice idea, just wanted to jot down some ideas while fresh:

Flag changes in direction above some threshold (90 deg?)
Threshold based approach: toggling between <20 | > 340 and >160 & < 200
some sort of expectation comparison, like predict the value of a point as being the midpoint of the point before and after it, and a point that is off by more than 100 deg is flagged

Bartdoekemeijer commented 1 year ago

Yeah good ideas! Maybe the expectation comparison can leverage the neighbouring turbines? E.g., if the median change in wind direction of all turbines is 10 deg, but there are one or two turbines reporting 150 degrees, then those are probably outliers.

paulf81 commented 1 year ago

That also makes sense to me, could be a threshold like, can't exceed some X times the median change?