3. ChargEval¶

The electric vehicle infrastructure decision support system (ChargEval) is the tool to guide decision makers in planning the deployment of EV infrastructure. Public agencies as well as private companies can use the ChargEval, since it predicts KPIs relevant to each domain. The ChargEval has been developed in collaboration with and with funding from Washington State Department of Transportation. The results and scenarios presented in this paper are therefore for the Washington state region. However, the methodology described is general enough to be applied anywhere across the world.

3.1. Goals of ChargEval¶

The ChargEval shows the infeasibility metric across the state and indication of where a high of volume of traffic flow is matched with a high spacing of charging stations. These areas are good candidates for charging stations if the objective is to maximize utilization. Once the sites are chosen, the ChargEval estimates the potential utilization for a chosen site. The utilization is outputted as energy and power. The total energy consumed per charging station in a month and the peak power draw in a month are important statistics that affect the economics of charging stations.

3.2. Components¶

The various components of the ChargEval are described below. For an in-depth technical exposition refer to the corresponding paper or contact the authors.

3.2.1. Long Distance Travel Demand Model¶

The necessity of charging on a route is directly proportional to the number of EV trips passing through the route. The trip counts between origin-destination (OD) pairs were estimated using INRIX data, as reported in previous work. The OD matrix is composed of around 300k+ rows for indicating trip counts from all origin zips to all destination zips within WA. The trip counts are dependent on several factors like the origin and destination population, respective counties, and distance between origin and destination. The total trip count between an OD pair is composed of two quantities, the traffic belonging to O, departing from O to D, and the traffic belonging to D, returning to D from O. It is imperative to separate the total traffic volume between an OD pair into returning and departing sub-volumes as the sub-volumes are dependent on the county that the EV belongs to. For techinal details refer to this publication.

3.2.2. Vehicle Choice Decision Model¶

The vehicle choice decision model (VCDM) can tell us whether an EV is feasible for trip depending on various trip and vehicle characteristics. Ge estimates several discrete choice models generated through stated preference surveys. For the purpose of EVI-ABM a latent choice logistic regression model is used that makes a vehicle selection between an internal combustion engine vehicle (ICEV), a rental vehicle or a battery electric vehicle (BEV), and is described as under:

(3.1)¶\[u_{\text{icev}_i} = \theta_\text{1} \times \text{gas cost}_\text{i,icev} + \varepsilon_{\text{icev}_i}\]

(3.2)¶\[u_{\text{rent}_i} = \theta_2 \times C_{\text{rental}_i} + \theta_3 \times \text{gas cost}_\text{i,rent} + \varepsilon_{\text{rent}_i}\]

(3.3)¶\[u_{\text{bev}_i} = \theta_4 \times \frac { L }{ r_\text{full}} + \theta_5 \times \frac { \text{Max}_\text{Spacing} }{ r_\text{full} } + \theta_6 \times l_\text{restrooms} + \theta_7 \times \text{Restaurants} + \theta_8 \times \text{Des}_{\text{charger}_\text{type(L2)}} + \theta_9 \times \text{Des}_{\text{charger}_\text{type(L3)}} + \text{ASC_BEV} + \varepsilon_{\text{rent}_i}\]

In the above equations, $u$ represents the utility of the particular vehicle choice. $\theta_i$ are the model coefficients for the covariates: cost of gas for ICEV during the trip ($\text{gas cost}_\text{i,icev}$), cost of a rental car ($C_{\text{rental}_i}$), gas cost for a rental car ($\text{gas cost}_\text{i,rent}$), ratio of trip length and full range of BEV ($\frac { L }{ r_\text{full}}$), ratio of maximum spacing between chargers along the trip route and full range of a BEV (${ \text{Max}_\text{Spacing} }{ r_\text{full} }$), largest spacing between restrooms along the route ($l_\text{restrooms}$), whether there is a restroom near the charging station ($\text{Restaurants}$), whether the destination has a level-2 charger ($\text{Des}_{\text{charger}_\text{type(L2)}}$), whether the destination has a fast charger ($\text{Des}_{\text{charger}_\text{type(L3)}}$). $\text{ASC_BEV}$ is the alternative specific constant for BEV and $\varepsilon$ are the error terms. The coefficients for the variables used in this study are presented in Table 1.

Table 1: Vehicle Choice Decision Model Parameter Estimates

Covariates	Estimate	P-value
ICEV gas cost ($) $\theta_1$	-0.040	0.000
RENT cost ($) $\theta_2$	0.059	0.010
RENT gas cost ($) $\theta_3$	-0.075	0.000
relative distance $\theta_4$	-1.659	0.002
relative max spacing $\theta_5$	-9.342	0.000
furthest restroom break (miles) $\theta_6$	0.002	0.271
Restaurants $\theta_7$	0.197	0.688
Des charger (Level 2) $\theta_8$	-0.748	0.141
Des charger (Level 3) $\theta_9$	1.428	0.039
$\text{ASC_BEV}$	11.184	0.000

3.2.3. EV Infrastructure Agent-based Model (eviabm)¶

EV Infrastructure Agent-based Model (eviabm), is an agent-based model for modeling the utilization of EVSE in the state of Washington. As such, it has the following attributes:

Agents:

Electric vehicles in the state of WA: We consider all the electric vehicles registered in the state of WA as our EV agents. While some EVs maybe travelling outside the state and some out of state vehicles maybe traveling within WA, for the present study, we ignore these vehicles. Source: Washington State Department of Licensing.
Washington road network: The EVs move on roads and travel is restricted to roads. Currently, we ignore the elevation of the roads, but in future, the roadway elevation can be included, and the energy model can account for the changes in elevation. Source: Washington State Department of Transportation.
Electric Vehicle Supply Equipment / Charging Stations: The charging stations are the agents where the EVs charge when they are charge depleted. The instantaneous power drawn and total energy consumed are the EVSE utilization outputs from the simulation that we are interested in. Source: Alternative Fuels Data Center.

Environment: Currently, a two-dimensional simulation is bounded by the state of WA.
Time: A single simulation runs for 24 hours in 1-minute time-steps. This means that we simulate EV travel around the state for a period of one day at a time and update the states of our agents each minute.

3.2.4. Charging Choice Decision Model¶

While the vehicle in enroute its destination, it might need to charge along the way. The choice of charging at a charging station can modeled by a decision choice model. Among the various models developed by Ge, we use the static choice decision model. The model equations are as under:

(3.4)¶\[u_{\text{charging}_\text{it}} = \theta_0 + \theta_1 \times \text{SOC}_\text{it} + \theta_2 \times \text{DEV}_\text{it} + \theta_3 \times \text{Hours} + \theta_4 \times C_{\text{charging}_\text{it}} + \theta_5 \times T_{\text{charging}_\text{it}} + \theta_6 \times T_{\text{access}_\text{it}} + \theta_7 \times \text{Amenity}_{\text{restroom}_\text{it}} + \theta_8 \times \text{Amenity}_{\text{more}_\text{it}} + \varepsilon_{\text{charging}_\text{it}}\]

In (3.4), $u$ represents the utility of charging, $\theta_i$ are the model coefficients, $SOC$ represents the state of charge of the BEV, $DEV$ is a Boolean denoting whether the vehicle has enough range to reach the next charger if it chooses to not charge at this charger, $Hours$ represents the hours the driver has been driving the vehicle, $C_\text{charging}$ represents the cost of charging the vehicle, $T_\text{charging}$ refers to the time taken to charge the vehicle, $T_\text{access}$ represents the time taken to access the charging station from the current route, $\text{Amenity}_\text{restroom}$ represents whether we have restroom as an amenity at the location of charging station, $\text{Amenity}_\text{more}$ represents whether we have more amenities like restaurants, Wi-Fi at the charging station location, and $\varepsilon_\text{charging}$ represents the error. The coefficients for the charging choice decision model used are as presented in Table 2.

Table 2: Charging Choice Decision Model Parameter Estimates

Covariates	Estimate
(Intercept) $\theta_0$	2.034***
SOC (%) $\theta_1$	-4.584***
Deviation (DEV) $\theta_2$	2.440***
Time in car (h) (Hours) $\theta_3$	-0.069
Charging cost ($) ( $C_\text{charging}$ ) $\theta_4$	-0.010***
Charging time (h) ( $T_\text{charging}$ ) $\theta_5$	-0.242**
Access time (min) ( $T_\text{access}$ ) $\theta_6$	-0.025***
Amenity: restroom only ( $\text{Amenity}_\text{restroom}$ ) $\theta_7$	0.049
Amenity: restroom, dining & WIFI ( $\text{Amenity}_\text{more}$ ) $\theta_8$	0.213**

* p-value < 0.1; ** p-value < 0.05; *** p-value < 0.01