Yes, the chief engineer did mention that they measure 96 voltages, and that the lowest / highest was part of the equation, as we know it has to be.
Here is a crack at the complexity of challenges the Nissan algorithm programmers have to rise to.
Estimated Distance to Empty (DTE)
Energy consumption ratio could be more varible factor than SOC.
DTE would be better indicator - carefule drivers
DTE(km)=SOC(KWH) x Energy consumption rate (km/KWH)
Changed by Acceleration, Regen, Climae Ctrl, Gradient etc.
Here is my more simplified take-away from Kedota-san's last chart together with my own experience using the SOC meter.
I don't believe there is any disagreement that in the DTE equation above, most of the uncertainty in DTE derives from the Energy consumption rate factor, not the SOC. However, on this graph Kedota-san was illustrating that the SOC is not perfect.
The accuracy of the internally-generated SOC estimate (gids) is complicated by residual cell imbalance, temperature, rate of discharge, etc. Dispite this complexity, I feel the estimate presently generated is impressively accurate. At the very least it is stable
. Yes, it does exhibit artifacts, such as sometimes going down at the beginning of regen, or up after the car sits off for a few minutes, but these artifacts are almost always +- 1 gid, which represents only about .3% of the total. I agree with a comment by Tony Williams that the SOC algorithm probably uses something like coloumb counting while discharging/charging recalibrated by OCV (open circuit voltage) when the battery has been idle for a short period.
The key point I saw on Kedota-san's graph was the importance both of cell-imbalance and temperature. There were no actual units on the graph, but it was clear how, for example only, a 1% variation of voltage over all 96 cell pairs would cause as much as 2% loss of battery capacity: 1% at the top of charge and 1% at the bottom.
I am seeing the temperature effect in the rate my Leaf charges. In warmer weather at the end of summer it was taking 4 minutes per 1% charge in the 80-90% range, and now it is taking only 3.5 minutes per 1% charge. Since I believe the power going into the battery is the same (I have not verified this), this implies a temporary reduction in pack capacity. It would be nice to get more precise data from others during this Winter.
I heard one suggestion during the discussion concerning the "predict the future" problem with the consumption factor: someone suggested that "crowd-sourcing" of Leaf Carwings data together with altitude profile data could yield better estimates of "consumption to destination." Perhaps a little unrealistic for now, but worth considering.