#evc john smart, david rempel (ergonomist), gustavo cezar ![[Pasted image 20230322122102.png]] ![[Pasted image 20230322122309.png]] - non of the four payments worked for 7.3% - ![[Pasted image 20230322122426.png]] - definition of uptime can be gamed![[Pasted image 20230322122542.png]] - most EVSE from installation, but not for maintained - less ![[Pasted image 20230322122936.png]] ![[Pasted image 20230322123054.png]] county level - aggregated ( In your first figure of "Modeling: overview" slide, could you please elaborate on how you choosed the resolution of space (e.g. county or state level)? Also what do you mean by timer control? I am interested in applying hierarchical Bayesian for parameter estimation but these time&space resolution and horizon was challenging. ![[Pasted image 20230322123159.png]] real charging session data!! ![[Pasted image 20230322123244.png]] clustered segments from EV data ![[Pasted image 20230322123339.png]] combination of home and residential charging ![[Pasted image 20230322123422.png]] can generate scenarios, 15% of adoption in 4 different scenarios ![[Pasted image 20230322123510.png]] modeled grid for 2035 ![[Pasted image 20230322123644.png]] from total demand perspective (low home scenario's peak is higher) - but we should look at the net ![[Pasted image 20230322123824.png]] the more ev adoption we have, lower excess non-fossil (can't slow down) ![[Pasted image 20230322123929.png]] - addressing "net" is importatnt - ![[Pasted image 20230322124031.png]] - ![[Pasted image 20230322124040.png]] - ![[Pasted image 20230322124047.png]] gretchen macht - user behavior is important for - hit rate, heavy hitters (less grid perspective than gustav's) ![[Pasted image 20230322124117.png]] ![[Pasted image 20230322124241.png]] arima model for what demand would have been ![[Pasted image 20230322124412.png]] ![[Pasted image 20230322124456.png]] 608 users (in rhode island) ![[Pasted image 20230322124714.png]] rural vs urban users (where they live and how they use have relation) ![[Pasted image 20230322125131.png]] behavioral patterns of 70k (residential vs public level 2) novel alg. chouda (dcfc vs other options ![[Pasted image 20230322125157.png]]![[Pasted image 20230322125242.png]] ![[Pasted image 20230322125350.png]] prior vs non-prior (29% deadline rushihng to charge)![[Pasted image 20230322125357.png]] shift toward ![[Pasted image 20230322125435.png]] Gretchen - Q1. Was your motivation to apply Bayesian approach because you had clear evidence of different prior distribution for each behavior (procrastinating vs non- procrastinating and anxious vs non-anxious)? - dead line rush model (procrastination) looking at different (both prior) chunk of behavior (fact that there is proportion user who use this way - may not get this) Q2. There were little percent difference output (3% and 1%) and i wonder whether you are satisfied with these result and have plans to develop further. I am not sure researchers/audience in EV would be ready to put in all this additional modeling efforts if there is little difference. -> separating user scenario itself can be important NEVI standard timing people charge is not the best time, timer control shift (typically - aggregate) (charging profile) - space resolution: available data (survey, ev adoption, google search on ), county level - don't always need to dive deep (based on assumption) - couple of assumption informed three big EVSE companies and which were not functioning, Q2. how did you elicit prior distribution? (did you used data? i.e. empirical bayesian approach) in terms? this justifies all the effort that went in. prior knowledge that you wanted to include for behavior) 30% are procrastinating and 20 ![[Pasted image 20230322125521.png]]