September 8, 2016

Two new articles on electric vehicle charging

A new issue of Transportation Research Record: Journal of the Transportation Research Board has just been published and contains two papers by Sustainable Transportation Lab researchers. The papers use different data sources, but both concern the decision of electric vehicle (EV) drivers whether to plug in or not in a given situation. Understanding this decision is key to predicting the utilization of EV charging infrastructure and to designing effective infrastructure systems. Manuscripts of both papers (without final journal editing and layout) are available without a paywall on our publications page.

The first paper , which I co-authored with Haixiao Yu, studies the charging decisions of people driving Toyota Prius plug-in hybrids. We used data collected from 125 instrumented prototypes before the vehicle’s market introduction. We found a couple of cool things in doing this work, mostly pertaining to methods for modeling charging choices:

1. We found that the amount of energy that a car can get by plugging in is a better predictor of the charging decision than just the state of charge of the battery or the amount of time the car will be parked. This makes sense intuitively, but had not been previously demonstrated empirically. What was more surprising was that the “utility” or benefit that people derive from plugging in depends non-linearly on the amount of energy. In general, being able to get a little bit of energy (up to about 30% of a full charge, or 1 kWh in these vehicles) provides a lot of utility, but getting more energy beyond that provides relatively little additional utility:
2. Care is needed when identifying charging opportunities. We studied whether people charged their car when they were stopped at a charging location. The hard part of this was identifying all the locations where charging is possible. We identified these locations in several different ways, and we found that our results were sensitive to how we inferred the locations where charging was possible.
3. We compared two different methods for modeling heterogeneity in preferences. Mixed logit is a widely used method, but we found that latent class logit provided a better fitting model than mixed logit. We also note that latent class models are more interpretable in the sense that they divide respondents into discrete market segments with characteristic sets of preferences.

I coauthored the second paper with Yuan Wen and David Keith. We used data from an online survey of battery electric vehicle (BEV) owners to model the decision of whether or not to charge. We again found that the latent class logit approach provided a better fitting model than the mixed logit approach. The latent class model revealed three modes of behavior among BEV drivers in our survey, each represented by a class:

Class 2 was the largest, representing 60% of our respondents. This class is, on average, willing to pay an extra $2.35 for an hour for Level 2 charging (6.6 kW), and an extra $7.85 for an hour of DC fast charging (50 kW), relative to Level 1.

Given the heterogeneous preferences of different EV owners, an important implication of these papers is that a variety of public charging options should continue to be provided.