-
Notifications
You must be signed in to change notification settings - Fork 0
Variables: location - improve estimation #6
Comments
May need to revisit current home/not-home coding - @raffertyparker seems to be seeing some fast charging 'at home'? |
Yes, issues with GPS data and code that have to do work on. |
Would it be better to just use the kW to define home vs not-home - if we assume home charging is < 7 kW would this work? It assumes no fast chargers at home... |
Nope. As an example, I use free 3.3 kW destination charges all the time, and also charge at 3.3 kW at home. Most of my not-home charging is at 3.3 kW rather than at rapid chargers. |
hmm. @raffertyparker maybe for now we should not try to distinguish between home & not-home - just between 'standard' and 'rapid' - unless @zenourn can add the inferred Location field back to the data? |
@dataknut I could export it with a huge number of NAs, and then it'd also have cases with it saying not-home when they are home and vice versa. There is definitely GPS position data which is bogus and just needs to be completely thrown out. Other issue is that many charging locations are under cover where don't get a GPS signal so need to infer based upon previous readings (i.e., last known-locations they weren't home, but were getting close to home, then we lost GPS data and they started charging so probably charging at home). Then cases where were at home based upon last session GPS, no new GPS location acquired, go on a trip, then start charging again. Back at home? At a public charger? In case where high kW can infer a public charger, but if just at 3.3 kW can't really say anything. So to do home vs not-home correctly is actually a full model in itself based upon longitudinal GPS measurements with other variables to help inference. |
Agreed.
This sounds like too much work for what we would gain from it. |
When the GPS signal is missing for a sample (>30% of time) currently the location (home, not home) is assumed to be same as the last non-missing GPS signal. This imputation procedure could be greatly improved by using a probabilistic model of location taking into account previous and future known locations, time, charging current, etc.
The text was updated successfully, but these errors were encountered: