See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269854065

Locating Electric Vehicle Charging Stations

Article in Transportation Research Record Journal of the Transportation Research Board · December 2013
DOI: 10.3141/2385-04

CITATIONS READS
160 4,648

3 authors, including:

T. Donna Chen Kara Kockelman

University of Virginia University of Texas at Austin
44 PUBLICATIONS 1,821 CITATIONS 334 PUBLICATIONS 22,417 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by T. Donna Chen on 25 August 2015.

The user has requested enhancement of the downloaded file.

 THE ELECTRIC VEHICLE CHARGING STATION LOCATION PROBLEM:
A PARKING-BASED ASSIGNMENT METHOD FOR SEATTLE

T. Donna Chen
The University of Texas at Austin
6.9E Cockrell Jr. Hall
Austin, TX 78712-1076
donna.chen@utexas.edu

Kara M. Kockelman
(Corresponding author)
Professor and William J. Murray Jr. Fellow
Department of Civil, Architectural and Environmental Engineering
The University of Texas at Austin
kkockelm@mail.utexas.edu
Phone: 512-471-0210

Moby Khan
Cambridge Systematics, Oakland, California
mobashwir@gmail.com
Word Count: 5123 plus 9 figures and tables = 7373 word-equivalents

Proceedings of the 92nd Annual Meeting of the Transportation Research Board in Washington DC,
January 2013, and forthcoming in Transportation Research Record

ABSTRACT

Access to electric vehicle (EV) charging stations will impact EV adoption rates, use decisions, electrified
mile shares, petroleum demand, and power consumption across times of day. This work uses parking
information from over 30,000 personal-trip records in the Puget Sound Regional Council’s 2006
household travel survey to determine public (non-residential) parking locations and durations. Regression
equations predict parking demand variables (total vehicle-hours per zone/neighborhood and parked-time
per vehicle-trip) as a function of site accessibility, local jobs and population densities, trip attributes, and
other variables available in most regions and travel surveys. Several of these variables are key inputs for a
mixed integer programming problem, developed here for optimal EV-charging-station location
assignments. The algorithm minimizes EV users’ station access costs while penalizing unmet demand.
This useful specification was used to determine top locations for installing a constrained number of
charging stations within 10 miles of Seattle’s downtown, showing how charging location schemes’ access
costs respond to parking demand and station locations. The models developed here are generalizable to
data sets available for most any region, and can be used to make more informed decisions on station
locations around the world.

Key words: Plug-in Electric Vehicles, Charging Stations, Location Assignment Problem, Mixed Integer
Programming
BACKGROUND

As electric vehicles (EVs) enter the market, there is rising demand for public charging stations.
Symbiotically, the demand for EVs is influenced by the availability of refueling infrastructure: “Without
infrastructure, the vehicle of the future will remain just that – the vehicle of the future” (1). Provision of
public charging stations can diminish owners’ (potential and actual) range anxieties (2), thus increasing
EV purchase and use decisions. Morrow et al. (3) showed how an EV-based transport system’s overall
cost can be reduced by providing more charging infrastructure instead of investing in bigger batteries
(with greater range). They estimated that the marginal cost of increasing a car’s all-electric range (AER)
from 10 miles to 40 miles is $8,268, and the cost of installing an additional level-2 commercial charging
station (including administrative and circuit installation costs, assuming 10 charge cords per facility) is
$18,520. While the EV charging station location problem is a very new topic area, some important strides
have been made in the past few years.

Wang et al. (4) created a numerical method for the layout of charging stations using a multi-objective
planning model. Accounting for charging station attributes, distribution of gas-station demands (rather
than parking decisions, as a proxy for charging demands), and power grid infrastructure, among other
variables, the researchers tested and verified their model using data from Chengdu, China. Sweda and
Klabjan (5) used an agent-based decision support system to identify patterns of residential EV ownership
and driving activities to determine strategic locations for new charging infrastructure, with the Chicago
region as a case study.

Most station location problems are based on existing optimization routines/heuristics. For example,
Worley et al. (6) formulated the problem of locating stations and optimal EV routings as a discrete integer
programming problem, based on the classic Vehicle Routing Problem (VRP). Ge et al. (7) proposed a
method based on grid partition using genetic algorithms. Their routine focuses on minimizing users' loss
or cost to access charging stations after zoning the planning area with a grid partition methodby choosing
the best location within each partition, to reflect traffic density and station capacity constraints (which
include charging power, efficiency, and number of chargers per station). Li et al (8) also used genetic
algorithms to identify top locations for charging infrastructure. Their method is based on conservation
theory of regional traffic flows, taking EVs within each district as fixed load points for charging stations.
The number and distribution of EVs are forecasted, and the cost-minimizing charging station problem is
(heuristically) solved using genetic algorithms.

Frade et al. (9) used Lisbon, Portugal as a case study, for application of a maximal covering location
model (10) to maximize the EV charging demand served by an acceptable level of service. They
determined not just the locations, but also the capacity of stations to be installed at each location. Finally,
Kameda and Mukai (11) developed an optimization routine for locating charging stations, relying on taxi
data and focusing on stations for Japan’s recently introduced on-demand bus system.

This paper adds to the growing field of charging station location solutions by providing behavioral
models to predict when and where vehicles are likely to be parked. It also takes a different approach for
anticipating charging demand, using parking demand as a proxy. The optimization routine used
recognizes parking demands across Seattle neighborhoods/zones, and ensures that stations are not too
clustered, by minimizing total system travel distances to the closest charging station, after assuming a
maximum cost for those parking beyond the limit of reasonable walk access. The mixed integer program
(MIP) developed here is based on the “fixed-charge facility location model” (12), which identifies a set of
facility sites to minimize the cost of serving a set of demands (located over space/across sites). This type
of model has been used to design communication networks (13), locate off-shore drilling platforms (14),
and locate freight distribution centers (15). Within the realm of EV charging station location work, the
closest parallel to this research is likely Hanabusa and Horiguchi’s (16) framework, minimizing EV travel
costs while maintaining a minimum buffer distance around each charging station. Their EV travel cost is
a function of travel time and waiting time at each charging station, and they define demand at each
charging station via a traffic assignment algorithm (based on route choice behavior). The current paper
attempts to best satisfy demand for public charging of EVs based on parking durations, land use
attributes, and (in the case of individual parking durations) trip characteristics. Optimal station locations
are determined as a function of parking demand and access (walk) distances/costs.

DATA DESCRIPTION

The data used for this project was obtained from the Puget Sound Regional Council’s (PSRC) 2006
household travel survey. The trip data contains trip information of 4,741 households and 10,510
individuals residing in the King, Kitsap, Pierce and Snohomish counties of Washington State. Each
respondent was asked to keep a travel diary for two consecutive days, all of which were weekdays. The
region consists of 3,700 Traffic Analysis Zones (TAZs), as shown in Figure 1.

The entire region consists of 1,177,140 parcels, and each trip in the trip file is connected to an origin and
destination parcel identification number or “ID”. Each household is also connected to its home-location
parcel ID. In contrast to other regions’ TAZ-based land use data sets, PSRC land-use information is
available for each parcel, and each trip in the Seattle data is associated with an origin parcel and a
destination parcel, along with parking, transit and land use attributes within quarter-mile and half-mile
buffers/radii around each of these parcels. Buffer-based variables include number of housing units,
numbers of jobs (by sector: education, food service, government, industrial, medical, office, construction,
retail and service), average costs of nearby parking (both hourly and daily), number of free off-street
parking spaces, number of intersections (by type: four-way, three-way, and “point” nodes/dead-ends),
number of local and express bus stops, (network) distance to nearest bus stop, and other variables. The
wealth and resolution of information provided makes this data set unusual and well suited for analyses
performed in this paper.

Table 1 shows various descriptive statistics for Seattle’s surveyed persons and households. The average
respondent is 42 years old, and 47% of respondents are males. 78 percent are licensed drivers, and the
average numbers of persons, workers and vehicles in each household are 2.22, 1.13 and 1.89,
respectively.

METHODOLOGY

In order to relate the Seattle region household travel survey data to optimal charging station locations for
parked EVs, this study took a three step approach. First, parking locations (by parcel, then aggregated by
TAZ) and durations were determined for all trips away from home and of at least 15 minutes in duration
(i.e., those that serve as plausible candidates for public charging, if an EV had been used). This parking
duration information was then used for regression models that relate (1) zone-level parking demands
(aggregated across sampled trips) to land use attributes and (2) trip-level parking demand to individual
trip characteristics. Parking demands were also used as inputs for identifying optimal charging station
locations, to satisfy as much demand as possible, subject to certain constraints (on access and station
supply). The formulation of a mixed-integer optimization problem is presented here, along with an
illustrative application to 900 TAZs near the region’s center.

Determining Parking Locations and Duration

Parcel-level parking information was extracted from the trip data file in order to determine where vehicles
are parked in the region and for how long. A snapshot of the trip data is shown in Table 2.
The trip data consist of 87,600 person-trips, but not all were by car or light truck. After eliminating all
passenger trips (in order to avoid duplicating driver trips) and keeping only trips made by light-duty
vehicles, 48,789 trips remained in the data set. To estimate public charging demand at different parcels
(and then at the level of TAZs), the following steps were used:
1. Consecutive trips were identified, where the destination of the earlier trip coincides with the
origin of the later trip. The time between these two trips is when a vehicle is parked at that unique
parcel.
2. No parking at one’s home parcel is counted, since parking locations at home are not of interest for
locating public charging stations.
3. Parking durations of less than 15 minutes were removed, since those are not enough for Level II
charging. (Level III charging stations would not have this restriction.)
A MATLAB script was written to perform the above analysis for all 48,789 trips. As a result, 30,085
candidate parking durations for public charging emerged. The output consists of tripmaker ID, parking
parcel ID, and parking duration. The parking information (at the parcel level) was then aggregated by
TAZ for neighborhood analyses of parking demand, as described below.
Forecasting Zone-Level Parking Demand

The demand for public EV charging in each TAZ may be roughly proportional to the total duration of
parking for all surveyed light-duty vehicles at that TAZ (outside of those that park at their home parcels).
This parking duration was first normalized, by dividing by parcel size, resulting in values of parking
duration per square mile, for each TAZ. Out of the 3700 TAZs in the Seattle region, eight did not contain
any parcels identified with land use attributes and were not used in the zone-based analyses. Summary
statistics for total surveyed parking durations and other variables of interest (as predictors of parking
demand) of the remaining 3692 TAZs are shown in Table 3.

Table 4 shows the parameter estimation results of an ordinary least squares (OLS) regression of parking
duration (per square mile) on covariates like population and jobs densities. All relevant land use, access,
and network connectivity variables were tested as covariates in initial regression model specifications,
with statistically significant regressors (at the .05 level) retained in the final model. Standardized
coefficients are also shown, to highlight levels of practical significance. These represent the number of
standard deviation (SD) changes in the response variable (parking duration per square mile) following a
one SD change in the associated covariate (evaluating all parameter values at their means).

Based on the model’s standardized coefficients, parking demand’s intensity (per square mile) is most
associated with employment (jobs) density. Parking prices and transit access are also relevant, but
secondary. Increased student density and network connectivity (via more four-way intersections) also
appear to play meaningful roles in increasing a zone’s total parking demand for the zone.

Forecasting Trip-Level Parking Durations

In addition to examining total parking demand per zone, parking durations for individual drivers/parked
vehicles, in minutes per destination, can be modeled as a function of trip and destination characteristics.
Table 5 presents summary statistics of these individual trip attributes (parked away from home, for at
least 15 minutes), along with average parking durations for various activity types/trip purposes. As can be
seen, work trips are the most common trip type (27 percent of the total) and command the longest parking
durations (among away-from-home parking experiences), averaging 380 minutes or 6.33 hours each.

Table 6 presents OLS regression estimates for these trips. All relevant zone characteristics (including
regressors tested in the zone-level parking demand model, such as parking prices, transit, and network
characteristics) and trip-level variables were tested as covariates in initial models, but only statistically
significant regressors (at the .05 level) were retained in the final model.

Table 6’s results offer a wide range of interesting results, with many very practically significant predictor
variables. Since individual trips provide a very large data set, t-statistics are high; fortunately, model fit is
also strong (R2adj = 0.590). Apparently, activity type at one’s destination is what most heavily influences
parking duration, with work trips and K-12 school trips having roughly equal and very long parking
durations, on average (for those who drive for such activities [very rare for K-12 trips, since most of these
students are not driving themselves to school]).

Longer parking durations are also evident for college, religious/community, recreational, and social
activities. Predictably, trips made for the purposes of picking up/dropping off passengers entail the
shortest parking durations, on average. Interestingly, job density is not statistically significant in this
model despite the fact that the correlation between work-trip purpose indicator and job density is low (ρ =
+0.11).

Trips involving passengers are predicted to require slightly shorter parking durations than single-
occupant-vehicle (SOV) trips, while longer-distance trips increase durations, as expected (by about 3.4
minutes per mile, everything else constant). Such information is useful for charging station owners and
operators, who will want to anticipate how many people can and will charge at a station or set of stations,
and for how long. Station availability, upon arrival of an EV, can be paramount for station success (by
encouraging further EV adoption and future EV trips to that station).

Anticipating Best Sites for Public Charging Stations

The modeling results discussed above illuminate a variety of factors that contribute to (or at least are
associated with) zone-level and trip-level parking demand, in statistically and practically significant ways.
In order to select highly accessible, high-demand spots for installation of public charging stations, an
optimization problem was specified, with an objective function that seeks to minimize the total access
costs (walk distances) from the charging stations to drivers’ ultimate destination zones (TAZs). Here, EV
charging demands are assumed proportional to (i.e., well proxied by) light-duty-vehicle parking demands,
as reported directly in the sample data. The optimization ensures a minimum distance between charging
stations, to avoid clustering of the not-inexpensive charging infrastructure in adjacent high-parking-
demand zones. Such problems are solved using mixed-integer programs (MIPs), which are common in
transportation applications, such as airline crew scheduling, vehicle routing, and pipeline design (e.g., 17,
18, 19). MIPs are generally solved using branch and bound techniques (20).

The following set of equations defines the problem solved here using the General Algebraic Modeling
System (GAMS), a software designed for mathematical programming and optimization tailored for large-
scale modeling applications. Outside of proprietary programs, noncommercial freeware is also available
for solving MIPs, such as ABACUS and bonsaiG (20).

The Mixed-Integer Optimization Problem

The objective function in this MIP (shown below) aims to reduce total access cost as a function of walk
distance between zones and ( ) weighted by parking duration. The walk penalty is limited to a
maximum distance or cost, (set to 2 miles in this application), since drivers are unlikely to walk long
distances for parking (similar to transit-access experiences [21]). Here i and j index the set of zones for all
potential destination TAZs and assignment of individual charging stations, respectively. In this set up, the
number of charging stations ( ) is less than the number of zones ( ) due to budget constraints. Then, for
EVs whose destination is some zone not equipped with a charging station, the parking demand will have
to be satisfied by a (hopefully close by) charging station in zone . In other words, represents the
parking demand for zone met by a charging station in zone . Assuming that overall parking demand is
likely proportional to EV parking demand, the objective function penalizes longer parking access
distances proportional to parking demand.

In addition to , another key decision variable is , which takes on a value of 1 for zones with charging
stations and zero for zones without, representing the set of optimal charging location zones. Other
parameters include , the parking demand at zone , and L, which is the limit on the total number of
charging stations one can allocate to zones. To ensure that charging stations are sufficiently spaced out,
the indicator takes on a value of 1 if the distance between and is less than a specified minimum
spacing and zero otherwise. A large number allows all parking demand to be assigned to charging
stations, hopefully ensuring that locally parked EVs can be accommodated by their nearest charging
station.

Objective function:
min ∑ ∑

Constraints:
1. ∑ = , ∀ ∈ (parking demand constraint)
2. ∑ ≤ , ∀ ∈ (charging supply constraint)
3. ∑ ≤ , ∀ ∈ (charging-station availability constraint)
4. ∑ ≤ 1, ∀ ∈ (charging station spacing constraint)
5. ≥ 0 ∀ ∈ , ∈ (non-negativity constraint on parking demand)
6. ∈ 0,1 ∀ ∈ (binary variable constraint for charging station selection)
1 <
7. = (minimum inter-station spacing)
0 otherwise
8. ≤ (maximum access cost)

The formulation of this optimization problem also introduces some challenges. Capacity for each
charging station is undefined here, since parking demand is currently without a time-of-day dimension
and the objective function may overly favor work and school trips, with their long parking durations.
Here, the optimization simply aims to locate optimal zones with a reasonable spread under the assumption
that EV parking patterns will imitate overall parking demand. Nonetheless, the MIP specified here is a
step towards efficiency of locating charging stations, as illustrated in the Seattle application below.

Charging Station Allocation: A Seattle Area Application

To demonstrate the mixed-integer optimization problem, total daily parking demand in minutes ( )
across 900 TAZs (i = 900, j = 900) within approximately 10 miles (network distance) of the Seattle CBD
were considered. (Inclusion of all 3,700 zones resulted in a large matrix that caused the GAMS software
to time out in the search for a solution.) With relatively small size (just 5 percent area of the average
PSRC TAZ) and high population density (three time that of the average PSRC zone), these relatively
central TAZs are good candidates for inter-zonal parking access of EV charging stations. (Large, low-
density peripheral zones cause problems for the GAMS algorithm because they have no or few
neighboring zones within the 2-mile maximum parking-access distance. For such applications, the zones
with larger areas can be split, with all zones of approximate equal size being the optimal condition for
seeking a set of solutions.)
The total number of charging stations was limited to L = 80, and the minimum distance between charging
stations was set to = 1 mile, since ½-mile access/walk distances (the worst-case scenario for an EV
owner parked between such stations) are often reasonable, especially for workers intending to be parked
for many hours. Network walking distances (as given by the PSRC, which exclude freeway links and
certain bridges considered unsafe for pedestrian use) were used to represent the travel costs, . As noted
earlier, the maximum walk penalty, , was limited to 2 miles, to reflect a cap on reasonable access costs.
Using the Coin- or Branch-and-Cut (CBC) solver, a straightforward GAMS code inputted the travel cost
matrix. To reduce the GAMS-required memory, the large travel cost matrix was filtered to restrict parking
assignment to charging stations within a 2 mile access distance. The algorithm arrived at a solution in
approximately 8 minutes and 45 seconds on a standard desktop computer. The mixed-integer problem
selected optimal parking station locations in the 80 zones listed in Table 7 (and shown in Figure 2), with
PSRC travel survey parking demands and associated zone ranks (based simply on that demand) shown
alongside.

As shown in Table 7, many of these zones rank high in parking demand, so a charging station scores well
by serving them directly. But many others were also selected, despite very low in-zone parking demand,
thanks to their strategic locations – nestled among other zones with high parking demands. Figure 2
provides maps of this solution set for charging station locations (left side) versus a simple assignment
approach, where chargers are placed in the 80 zones with highest parking demands (right). As illustrated
in Figure 2, optimal station locations are much more scattered throughout the 900-zone region, versus the
simple demand-based assignment method, which concentrates stations in the region’s central business
district. When clustered together, high parking demand zones, under the optimal solution, are sometimes
served by a low- to medium-demand zone, nestled among them.

The optimized solution yielded a total (minimized) cost (z) of 842,413 mile-minutes, with a weighted-
average parking access cost of 0.69 miles (weighted by total parking duration of station-assigned zones).
79.9 percent of parking demand was able to access a charging station within 1 mile of the destination
TAZ, with a maximum walk-access distance of 1.90 miles. In contrast, if charging stations were placed
purely based on parking demand (in the top 80 TAZs, where parking demand is highest), the total cost (z)
would be 890,135 mile-minutes (5.7 percent higher than the optimal solution found here), the average
parking access cost would be 0.73 miles, and the maximum access distance would more than double, to
3.96 miles. Under this simplified approach, 78.0 percent of parking demand appears able to access a
charging station within 1 mile of the destination TAZ.

Such results suggests there is some merit to simple, demand-based assignment (with no more than one
station per zone, though that station may have multiple chargers available). However, for implementations
where zones are even smaller in size (with greater opportunity for interzonal parking access), the benefits
of this paper’s optimization approach are more striking. For example, when 20 charging stations are
strategically located across the City of Seattle’s 218 zones, the routine returns an optimal solution in
under 1 second, placing 94.5 percent of parking demand within 1 mile a station – rather than meeting just
79.6 percent of parking demand (within 1 mile) under the simple assignment rule (allocating public
chargers to zones with highest parking demands).

CONCLUSIONS

A key factor for long-term EV success involves simplifying the logistics of charging one’s vehicle away
from home. Thoughtful siting of public charging stations can ease consumer range anxiety while offering
a lower cost approach to integrating EVs into the transportation market (versus investing in longer-range
batteries). This study relied on household travel survey data from the Seattle region to investigate parking
demands (by zone and by trip) and then identify optimal station locations using a rigorous MIP.
Parking demand was examined in two ways, based on land-use characteristics for a zone and trip (and
traveler) characteristics for individual trips. Land use and access attributes were used in an OLS
regression model to predict total parking times per zone. Parking demand (per square mile) at the TAZ
level rose significantly with jobs and student densities. More connected and transit-served zones,
characterized by more nearby four-way intersections and bus stops, were also found to experience higher
parking demand. At the trip level, trip purposes were by far the most significant predictor of parking
durations. Models revealed that work and school trips require the longest parking periods while regular
errands (personal business, shopping, eating out, and picking up and dropping off passengers)
necessitated the shortest parking durations, with social and recreational activities falling somewhere in
between. Trip distance and use of a car (rather another vehicle type) also lengthened average parking
durations.

The first regression model’s outputs provide key inputs for determining efficient charging station
locations, as specified here via a mixed integer optimization program. Taking into consideration budget
constraints (which limit total number of charging stations to be deployed), and avoiding resource
clustering (by specifying minimum station spacings), the optimization problem assigned 80 public
charging stations thoughtfully across 900 TAZs within 10 miles of Seattle’s downtown center. As
designed, these were spaced at least one mile apart, with wide ranging access and parking demand
characteristics, illustrating both the importance of parking intensity and access. This optimal charging
location scheme was compared to one based focused on top-ranked zones, in terms of parking demand,
and yielded clearly better results in multiple ways, like average access distances (in addition to
minimizing total access costs).

The work presented here has certain limitations. It assumes that LDV parking demand is a strong proxy
for EV charging demands, which may not reflect actual charging demands, particularly while EV market
shares are still small. As compared to the general U.S. population, early EV adopters are
disproportionately younger, male, more educated, and more environmentally sensitive (22). Over time, as
EV market shares grow, parking demands may more closely reflect EV charging demands. Introducing a
time-of-day dimension to the optimization problem, to reflect the dynamic nature of charging demand
levels, would also serve as a useful extension. While this work’s MIP identifies optimal zones for
charging station placement, specific station locations within identified zones are not defined. Such
location choices are likely to be highly influenced by visibility, accessibility, and installation costs, which
vary from $2000 to $5000 for wall-mounted stations in parking garages to $15,000 or more for stations
which require utility service and infrastructure upgrades, according to Austin Energy staff (23).

Nonetheless, the models developed here provide a basic framework for readers to anticipate parking
demands and more efficiently locate EV charging infrastructure in new settings and/or subject to different
constraints (on access costs and station availability). This framework can be quickly adapted to other
cities and regions, with similar data sets, for making more optimal decisions on station locations around
the world.

ACKNOWLEDGEMENTS

The authors are grateful to the National Science Foundation for financial sponsorship of this work
through the University of Texas at Austin and Texas A&M’s Industry-University Research
Center for PEVs. Crucial support in the form of optimization formulation aid by Tarun Rambha, GAMS
guidance by Christopher Melson and Dr. David Morton, and administrative assistance by Annette Perrone
also helped make this investigation possible. The suggestions of several anonymous reviewers are also
appreciated.
REFERENCES

1. Wehrman, J. (2013). Senior vice president of Honda of America Manufacturing. As quoted by A.

Lienert in Alt-Fuel Vehicle Infrastructure is Stumbling Block, Honda Tells Lawmakers.
Edmunds.com, March 6, 2013. Article available online at http://www.edmunds.com/car-news/alt-
fuel-vehicle-infrastructure-is-stumbling-block-honda-tells-lawmakers.html
2. Aoki, H. (2010). The Standardization of Electric Mobility Towards CO2 Reduction. Slideshow
developed for The Tokyo Electric Power Company.
3. Morrow, K., Karner, D. and Francfort, J. (2008) Plug-in hybrid Electric Vehicle Charging
Infrastructure. Final Report Idaho National Laboratory. INL/EXT-08–15058.
4. Wang, H., Huang, Q., Zhang, C. and Xia, A. (2010) A Novel Approach for the Layout of Electric
Vehicle Charging Station. Apperceiving Computing and Intelligence Analysis Conference.
5. Sweda, T. and Klabjan, D. (2011) An Agent-Based Decision Support System for Electric Vehicle
Charging Infrastructure Deployment. 7th IEEE Vehicle Power and Propulsion Conference, Chicago,
Illinois.
6. Worley, O., Klabjan, D. and Sweda, T. (2012) Simultaneous Vehicle Routing and Charging Station
Siting for Commercial Electric Vehicles. IEEE International Electric Vehicle Conference, Greenville,
NC.
7. Ge, S., Feng, L. and Liu, H. (2011) The Planning of Electric Vehicle Charging Station Based on Grid
Partition Method. IEEE Electrical and Control Engineering Conference, Yichang, China.
8. Li, Y., Li., L, Yong, J., Yao, Y. and Li, Z. (2011) Layout Planning of Electrical Vehicle Charging
Stations Based on Genetic Algorithm. Lecture Notes in Electrical Engineering, 1, Volume 99,
Electrical Power Systems and Computers: 661- 668.
9. Frade, I., Ribeiro, A., Goncalves, G. and Antunes, A.P. (2011) Optimal Location of Charging Stations
for Electric Vehicles in a Neighborhood in Lisbon, Portugal. In Transportation Research Record:
Journal of the Transportation Research Board, No. 2252: 91-98.
10. Church, R. L., and ReVelle, C. (1974) The Maximal Covering Location Problem. Papers of the
Regional Science Association (32): 101-118.
11. Kameda, H. and Mukai, N. (2011) Optimization of Charging Station Placement by Using Taxi Probe
Data for On-Demand Electrical Bus System. Lecture Notes in Computer Science 6883: 606-615.
12. Drezner, Z. and Hamacher, H.W. (Eds.). (2004) Facility Location: Application and Theory (2nd ed.).
New York: Springer.
13. Mirzain, A. (1985) Lagrangian Relaxation for the Star-Star Concentrator Location Problem:
Approximation Algorithm and Bounds. Networks 15: 1-20.
14. Balas, E. and Padberg, M.W. (1976). Set Partitioning: A Survey. SIAM Review 18: 710-760.
15. Nozick, L.K. and Turnquist, M.A. (1998) Integrating Inventory Impacts into a Fixed Charge Model
for Locating Distribution Centers. Transportation Research Part E 31(3): 173-186.
16. Hanabusa, H. and Horiguchi, R. (2011) Lecture Notes in Computer Science in Knowledge-based and
Intelligent Information and Engineering Systems, Volume 6883/2011, pp. 596-605.
17. Brimberg, J., P. Hansen, K.W. Lih, N. Mladenović and M. Breton (2003) An Oil Pipeline Design
Problem. Operations Research 51(2): 228-239.
18. Barnhart, C., A. Cohn, E. Johnson, D. Klabjan, G. Nemhauser and P. Vance (2003) Airline Crew
Scheduling, chapter 14 in Handbook of Transportation Science: International Series in Operations
Research & Management Science 56 (5): 517-560.
19. Kaufman, D., J. Nonisb, R. Smith (1998) A Mixed-Integer Linear Programming Model for Dynamic
Route Guidance. Transportation Research Part B 32 (6): 431-440.
20. Linderoth, J. and T. Ralphs (2004) Noncommercial Software for Mixed-Integer Linear Programming.
Technical Report 04T-023, Department of Industrial and Systems Engineering, Lehigh University.
Available at http://homepages.cae.wisc.edu/~linderot/papers/Linderoth-Ralphs-05-TR.pdf.
21. Transportation Research Board (2003). Transit Capacity and Quality of Service Manual, 2nd Edition.
Transit Cooperative Research Program (TCRP) Report 100.
22. Carley, S.R., R.M. Krause, B.W. Lane and J.D. Graham (2013) Intent to Purchase a Plug-In Electric
Vehicle: A Survey of Early Impressions in Large American Cities. Transportation Research Part
D18 (1): 39-45.
23. Freberg, Cameron (2012). Conservation Program Associate, Electric Vehicles and Emerging
Technologies, Austin Energy. Email to Kara Kockelman. November 9.
LIST OF TABLES
TABLE 1 Summary Statistics of PSRC Person- and Household-level Attributes
TABLE 2 Snapshot of Trip Data
TABLE 3 Summary Statistics of PSRC Zone Attributes
TABLE 4 Parking Demand (min/mile2) Regression Results (OLS)
TABLE 5 Summary Statistics of PSRC Trip Attributes
TABLE 6 Individual Parking Durations (min/trip) Regression Results (OLS)
TABLE 7 Charging Station Assignments and their In-Zone Parking Demands

LIST OF FIGURES
FIGURE 1 Map of the Seattle Region’s 3,700 Traffic Analysis Zones
FIGURE 2 Map of Optimal Charging Station Locations in Seattle vs. Top Parking Demand Zone
Charging Station Locations (I & J = 900, L=80)
Table 1: Summary Statistics of PSRC Person- and Household-level Attributes
Mean St Dev. Min Max
Person Records (N=10,510)
Age (years) 41.9 21.8 0 99
Male Indicator 0.47 0.50 0 1
Driver’s License Indicator 0.78 0.42 0 1
Student Indicator 0.21 0.4 0 1
Household Records (N=4,741)
Household Size 2.22 1.21 1 8
Household Number of Workers 1.13 0.85 0 5
Household Number of Vehicles 1.89 1.07 0 10
Number of Licensed Drivers 1.69 0.73 0 5
Household Income ($/year) 71,400 42,300 5,000 175,000
 Table 2: Snapshot of Trip Data

Household Person Trip Begin End Home

# # # Time Time Origin parcel Destin. parcel parcel
2045 1 1 1040 1100 8009 12543 12543
2045 1 2 1420 1450 12543 4532 12543
2045 1 3 2000 2055 4532 12543 12543
2045 2 1 810 900 12543 10093 12543
2045 2 2 1500 1540 10093 12543 12543
 Table 3: Summary Statistics of PSRC Zone Attributes
Variable (n = 3962) Mean Std. Dev. Min Max
2
Parking duration (mins/mile ) 2.41E+04 1.18E+05 0 2.43E+06
Population density (persons/mile2) 7.99E+03 1.88E+04 0 2.91E+05
Employment density (jobs/mile2) 1.26E+04 8.27E+04 0 2.07E+06
Student density (students/mile2) 1.64E+03 2.48E+04 0 1.02E+06
Housing density (units/mile2) 3.43E+03 8.20E+04 0 1.27E+05
Average price of daily paid parking within
zone ($) 0.145 1.027 0 21.3
Average price of hourly paid parking
within zone ($) 0.066 0.465 0 11.0
3-way intersections (1/2 mile radius) 45.8 22.52 0 119.3
4-way intersections (1/2 mile radius) 36.8 45.91 0 251.8
Express bus stops (1/4 mile radius) 2.25 6.369 0 55.6
Bus stops (1/4 mile radius) 6.21 9.016 0 69.6
Note: The parking duration is of surveyed vehicles only, which represent approximately 0.29 percent of
Seattle’s household-owned and operated light-duty fleet over a two-weekday period.
 Table 4: Parking Demand (min/mile2) Regression Results (OLS)

Parameter Standardized
Variable t-stat
Estimate Coef.
Constant 3268 1.06
Density
Population density (residents/mile2) -0.294 -0.047 -3.50
Employment density (jobs/mile2) 0.583 0.408 27.0
Student density (students/mile2) 0.226 0.047 4.11
Parking Prices (within ¼ mile)
Average price of daily paid parking ($) 2.22 0.193 11.0
Transit Access & Network Connectivity
#3-way intersections (within ½ mile) -158.0 -0.030 -2.41
#4-way intersections (within ½ mile) 160.8 0.062 2.94
#Express bus stops (within ¼ mile) 1537 0.083 3.29
#Bus stops (within ¼ mile) 1624 0.124 4.17
Number of Observations 3,692 TAZs
Adjusted R-squared 0.521
Note: All coefficients shown are statistically significant at the 5-percent level (p-value < 0.05). Y is the
zone’s total parking duration of surveyed drivers (away from home, and longer than 15 min duration).
Other covariates tested in Table 4’s model are all land use, network, pricing, and transit attributes shown in
Table 3.
 Table 5: Summary Statistics of PSRC Trip Attributes
Avg Parking
Duration
Variable Mean Std. Dev. Min Max (min. per trip)
Parking duration (min/trip) 142.0 199.5 15.0 2120 -
Trip distance (miles) 6.71 7.14 0.230 67.6 -
Passengers (excluding driver) 0.421 0.811 0 6 -
Activity: Work 0.271 0.445 0 1 379.7
Activity: School (K-12) 6.87E-03 0.083 0 1 338.8
Activity: College 7.63E-03 0.087 0 1 222.5
Activity: Eating out 0.071 0.257 0 1 46.1
Activity: Personal business 0.179 0.384 0 1 46.8
Activity: Everyday shopping 0.168 0.374 0 1 27.7
Activity: Major shopping 0.016 0.127 0 1 47.6
Activity: Religious/community 0.019 0.138 0 1 116.8
Activity: Social 0.040 0.197 0 1 127.6
Activity: Recreation-participate 0.057 0.232 0 1 103.5
Activity: Recreation-watch 0.016 0.126 0 1 107.4
Activity: Accompany someone else 8.88E-03 0.094 0 1 58.8
Activity: Pick up/drop off 0.133 0.340 0 1 15.5
Activity: Turn around 4.52E-03 0.094 0 1 53.0
Vehicle: Car 0.560 0.496 0 1 147.2
Vehicle: SUV 0.194 0.395 0 1 133.2
Vehicle: Van 0.119 0.324 0 1 103.3
Vehicle: Truck 0.089 0.285 0 1 173.7
Vehicle: Other 0.034 0.181 0 1 145.1
Note: n=30,085. Only trips ending away from home, with origin and destination zones in the region and
parked durations exceeding 15 minutes, are included here.
 Table 6: Individual Parking Durations (min/trip) Regression Results (OLS)

Parameter Standard.
Variable Estimate Coef. t-stat
Constant 372.2 125.5
Destination TAZ Characteristics
Land-use entropy (balance or mix index) -42.63 -0.047 -11.9
Distance to CBD (miles) -0.204 -0.013 -3.34
Population density (per mile2) -9.075E-05 -0.008 -1.99
Employment density (per mile2) 8.173E-05 -0.048 12.1
Student density (per mile2) -3.189E-05 -0.010 -2.48
Trip Characteristics
Trip distance (miles) 3.461 0.124 31.8
Passengers (excluding driver) -2.715 -0.011 -2.68
Activity: Work (base case) - - -
Activity: School (K-12) -21.28 -0.009 -2.37
Activity: College -157.6 -0.069 -18.4
Activity: Eating out -306.6 -0.396 -95.0
Activity: Personal business -313.2 -0.603 -135.6
Activity: Everyday shopping -324.5 -0.609 -133.9
Activity: Major shopping -308.0 -0.196 -51.5
Activity: Religious/community -246.3 -0.170 -44.6
Activity: Social -241.9 -0.238 -60.7
Activity: Recreation-participate -259.7 -0.302 -75.6
Activity: Recreation-watch -254.2 -0.161 -41.8
Activity: Accompany someone else -298.9 -0.141 -37.2
Activity: Pick up/drop off -344.1 -0.587 -127.3
Activity: Turn around -303.8 -0.102 -27.5
Vehicle: Car (base case) - - -
Vehicle: Van -13.87 -0.023 -5.82
Vehicle: SUV -4.101 -0.008 -2.12
Vehicle: Truck* -1.367 -0.002 -0.51
Vehicle: Other -17.88 -0.016 -4.22
Number of Observations 30,085
Adjusted R-squared 0.590
Note: All coefficients shown are statistically significant at the 5 percent level (p-value < 0.05) except those shown
with an asterisk (*). Bolded standardized coefficients indicate the most practically significant of covariates.
Table 7: Charging Station Assignments and their In-Zone Parking Demands

Parking Demand
TAZs Assigned a Survey Parking
Rank
Station (ID #) Demand (minutes)
(out of 900 zones)
305 29266 1
873 26473 2
709 13972 7
808 11489 8
674 10184 13
70 8959 18
30 6449 33
878 5909 41
804 5311 47
351 5154 50
384 4305 70
735 4040 72
844 3587 83
859 3317 92
721 3314 93
815 3237 96
701 3097 105
387 2669 125
557 2517 137
101 2445 144
31 2311 153
728 2284 155
332 2120 170
670 2035 175
803 1985 179
632 1854 198
679 1798 204
702 1655 223
17 1513 244
53 1481 247
141 1481 247
117 1264 271
2 1249 274
864 1248 275
416 1209 283
872 1191 286
757 1099 301
331 1089 303
428 1084 306
42 1077 309
745 1076 310
828 1005 335
882 1005 335
783 996 337
675 983 341
834 839 373
246 823 379
719 796 387
270 773 392
56 749 398
339 722 403
93 678 410
884 677 411
446 631 425
781 605 432
199 603 433
348 593 435
769 537 448
591 496 457
665 417 486
371 401 488
753 385 494
676 361 497
381 325 516
84 310 521
888 310 521
298 200 581
809 198 585
692 145 621
811 135 631
717 126 644
65 118 653
12 115 654
169 61 702
176 18 750
623 18 750
329 15 754
290 0 761
754 0 761
893 0 761
Figure 1: Map of the Seattle Region’s 3,700 Traffic Analysis Zones
 Figure 2. Map of Optimal Charging Station Locations in Seattle vs. Top Parking Demand Zone
Charging Station Locations (I & J = 900, L=80)

View publication stats