Sie sind auf Seite 1von 11

Transportation Research Part D 63 (2018) 377–387

Contents lists available at ScienceDirect

Transportation Research Part D


journal homepage: www.elsevier.com/locate/trd

Electric vehicles in car sharing networks – Challenges and


T
simulation model analysis

Stefan Illgen , Michael Höck
Technical University TU Bergakademie Freiberg, Department of Business Administration, Schlossplatz 1, 09596 Freiberg, Germany

A R T IC LE I N F O ABS TRA CT

Keywords: In this paper, we examine the operation of electric vehicles in urban car sharing networks. After
Electric vehicles surveying strategic and operational differences and comparing them to gasoline-fueled cars, a
Car sharing simulation study was carried out. The proposed discrete event simulation tool covered important
Transportation operational characteristics of electric vehicles, including realistic charging routines. Different
Mobility
vehicle types were compared under various conditions and on multiple markets to determine
Simulation
their performance. The data obtained indicated the competitiveness of electric vehicles in car
sharing networks. Key success factors included advantageous relations between the market en-
vironment (e.g. electricity and fuel prices) and important characteristics of electric cars (e.g.
price and range).

1. Introduction

The role of electric vehicles (EVs) in prospective and sustainable mobility is relevant for all modes of transportation, including car
sharing (CS). In particular, urban centers with high population densities are currently aiming to reduce the number of conventionally
powered vehicles (Engel-Yan and Passmore, 2013). Along with other alternative modes of transportation with the potential to reduce
emissions and/or overall traffic volumes, emission-free vehicles have become the focus of considerable attention (Moradi and
Vagnoni, 2018). After the introduction of the first CS system in Switzerland in 1948, several commercial projects where established in
larger cities around the world. Surprisingly, even one of the first CS networks in the world (WITKAR, Netherlands, established 1974)
operated EVs. Due to technical limitations including a maximum vehicle speed of only 30 km/h, it could not satisfy customer de-
mands to a sufficient degree. However, CS proved its utility in complementing or even replacing individual car ownership as an
alternative mobility concept during this period fostered by the oil crises of the 1970s (Millard-Ball, 2005; Cepolina and Farina, 2012).
Vehicle sharing experienced steady growth during the 1990s, and was revolutionized by the advent of mobile internet services and
devices (Shaheen and Cohen, 2013). The flexibility and spontaneity that distinguishes CS in general from conventional car rental
could now be utilized directly by customers through smartphone devices.
Today, CS companies face ever-greater demands and constantly extend their areas of operation (Lindloff et al., 2014). It is evident
that the combination of CS and e-mobility offers a promising mode of individual inner-city transportation. Additionally, the overlap
of key characteristics of CS and e-mobility provides a perfect application for both (Abdelkafi et al., 2013). Both are preferably used for
urban mobility and offer the potential to decrease emissions in transportation. They also benefit from higher utilization decreasing
fixed cost of a vehicle for its users. However, according to the recent study of some 101 CS providers in Germany by Parzinger et al.
(2016), 66 do not offer EVs in their current fleet. This number mostly refers to smaller enterprises with fleet sizes of less than 20
vehicles. However, high growth rates in the number of utilized EVs by CS providers have been identified and confirmed by 70% of the


Corresponding author.
E-mail addresses: stefan.illgen@bwl.tu-freiberg.de (S. Illgen), michael.hoeck@bwl.tu-freiberg.de (M. Höck).

https://doi.org/10.1016/j.trd.2018.06.011

Available online 13 June 2018


1361-9209/ © 2018 Elsevier Ltd. All rights reserved.
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

larger CS enterprises that already provide electric vehicles (Parzinger et al., 2016).
Our work in this area is directed toward increasing the proportion of EVs in CS fleets in order to fully utilize the ecological
potential of electric vehicle car sharing (EVCS) with regard to inner-city emission reduction. In contrast to the research in this field
conducted to date (see Sections 2 and 3), we employ a simulation approach built around a realistic vehicle state-of-charge (SOC)
simulation in Section 4. This methodological approach offers the flexibility to model typical characteristics of EVs operation under
stochastically distributed demand and trip times. It allows us to overcome the common but unrealistic simplification of fully re-
charging a vehicle before it reenters the pool of available vehicles in a CS network. We analyze the results from this simulation tool in
Section 5 and aim to answer the following research question: Might the operation of EVs in any CS network compete with gasoline-
fueled cars? The influence of different market conditions and different vehicle types are a major part of the study. Accordingly, if
applied as part of a decision support system, our work represents a preliminary phase to the common approaches based on mixed
integer programming and profit optimization for entire networks that can be found in many recent articles.

2. Literature

According to the notable work of Abdelkafi et al. (2013), CS represents a business model that could benefit from the typical
characteristics of e-mobility. This includes a feasible use case for limited vehicle range and preferably application in business fleets
due to the higher vehicle cost. Additionally, a satisfying charger infrastructure is expected to grow faster in urban areas than in rural
areas (Kihm and Trommer, 2014). A specific business model for an EVCS network can be found in Luè et al. (2012). The authors
design a CS network including customers and operators as well as private and commercial providers and highlight its contribution to
sustainable transportation (Luè et al., 2012). CS is also highlighted as a perfect application for EVs by Fairley (2013). The report relies
on current success and advantages of EVs as well as practical usage patterns mostly from a customer’s point of view (Fairley, 2013). In
contrast, the operators perspective is covered in Xu et al. (2017). Also success factors for EVCS systems are evaluated. The results
underline the high importance of customer satisfaction and the need for appropriate planning of EVCS (Xu et al., 2017). While the
publications noted above focus on operational issues, the following provide insights into the technical aspects of EVs that make them
suitable for CS. The high technical relevance of EVs (e.g. regarding smart energy storage) for future mobility is outlined by Farid
(2017) and Fuentes et al. (2017). Moreover, Vervaeke and Calabrese (2015) understand EVCS as a service that is experiencing
constant growth, thus underlining the significance of our own work for future urban mobility. Along with technical and operational
aspects, also the readiness of users to embrace EVs plays an important role. A survey analyzing the reactions and experiences of EVCS
users is presented by Kim et al. (2015). Yoon et al. (2017) found no added willingness to pay for EVCS in comparison with con-
ventional powered cars. However, market penetration of EVCS also depends on external factors. Considerations such as government
subsidies for vehicle purchase or the development of fuel and electricity prices comprise some of the factors that could promote or
hinder the adoption of EVCS (Reining et al., 2014; Plötz et al., 2015). After giving a short review on the thematically framework
behind electric vehicle sharing, the following quantitative studies offer noteworthy findings determining our own research.
The modeling of EVCS is quite advanced because some findings from studies of conventionally fueled CS vehicles could be
transferred to EVCS. Recent research, however, has already been designed for EVCS. For review purposes and for a general in-
troduction to the modeling and demand issues associated with one-way CS and the related vehicle relocation problem, the interested
reader may be directed to Le Vine et al. (2014), Shaheen et al. (2015) and Gavalas et al. (2015). Furthermore, there are examples for
modeling EVCS networks that deal with a wide range of research questions. The strategically oriented work conducted by Boyacı
et al. (2015) addresses the location planning of network stations depending on shifts in demand. Operational aspects like SOC of EVs
are neglected. Longer vehicle downtimes due to battery charging and specific recharge operations are investigated by Weikl and
Bogenberger (2015), Li et al. (2016), Brandstätter et al. (2017) and He et al. (2017). However, all of these papers are based on profit
optimization through mixed integer programming, thus exhibiting weaknesses that result in limited real-world applicability. In order
to avoid unmanageable problem sizes or model complexities, it is assumed that cars always fully recharge after any operation.
Practices such as idling in-between trips without recharging or partly recharging until the next customer arrives are not considered,
nor is even freeing up a charger due to the arrival of a vehicle with a lower SOC. Consequently, battery capacities (which determine
vehicle ranges) do not bear any influence on model outcomes unless they limit the distance of a single trip – which is irrelevant in
short-term CS applications. Moreover, underlying costs and profits in the target functions are based on case study-specific conditions
in only one market. Recent research conducted by Yoon and Cherry (2018) aims to compare conventional vehicles with EVs in a
dedicated sharing network. Their fleet optimization model leads to the conclusion that longer operation times are required to offer
the same profitability as EVs at least within their case study. A fixed vehicle range is assumed and recharging is not part of the model
(Yoon and Cherry, 2018).
Avoiding any simplification in the vehicle recharging process – thus simulating the actual SOC for any car individually, as well as
performing a market- and case study-independent cost analysis – represents a considerable research gap. This gap is addressed in our
paper with respect to the role of conventionally and electrically powered vehicle types vis-à-vis the efficiency and profitability of CS
networks.

3. Background

3.1. Scope and forms of CS

The two most important characteristics of a CS network are their trip type and station policy. Most CS systems offered today are

378
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

station based. Free floating systems are currently offered only by some companies in larger cities for example located in Germany.
Two-way trip systems require that customers return vehicles to the pick-up location, one-way systems on the other hand allow
customers to drop vehicles off at any station or within the free floating operation zone (Shaheen et al., 2015). The station-based mode
seems to be most qualified for EVs. Free-floating systems also function, though with some preconditions (Kopp et al., 2015). Either a
large number of accessible charging points are available in the free-floating operating area, or some customers will be forced to park
the car at specific recharge locations depending on the SOC. However, linking free vehicles to discrete charging points during idle
phases transforms such systems into one-way, station-based networks, which lie within the scope of this article (Wielinski et al.,
2016). A distinction between one-way and two-way CS was avoided during our initial research into EVCS. In our study it is not
relevant if a vehicle is returned at the location of origin or at a different station. Instead the EV and its specific workload within a CS
network is in the focus. Operational differences such as the vehicle relocation problem are relevant for electric and conventionally
fueled vehicles in a similar manner. Further insight into users preference of a particular CS system besides EVs is given in the notably
work of Yoon et al. (2017). Finally, the form of peer-to-peer or private CS falls outside the scope of this paper even though it may be a
good application for EVs. The idea of sharing the costs of ownership with friends, neighbors or community members is not limited by
the type of car. However, this raises the private ownership decision problem of choosing either a conventionally fueled vehicle or EV
independently from any peer-to-peer CS intentions.

3.2. Challenges in using EVs for CS

Many countries have passed legislation to promote increases in the proportion of EVs on their roads, whether on a voluntary or
compulsory basis. Today, meeting national requirements is often realized through the adjustment of commercial fleets (Reining et al.,
2014). CS comprises a positive application in this context. However, some general issues related to EVs appear in CS too. It is mostly
the technical limitations of EVs – such as their operating range or long charge times – that present challenging problems. Even though
technical progress has led to convincing increases in range, their performance under real-world conditions remains limited – e.g. at
low temperatures (Yuksel and Michalek, 2015; Zhang et al., 2017). Another technical aspect relevant for CS is that of battery
deterioration. Even a relatively young EV may have a low resale value due to the limited lifetime of its battery (Mahmoudzadeh
Andwari et al., 2017; Pelletier et al., 2017). However, frequent fleet rotation (which aids in securing acceptable vehicle resale values)
is an important factor for CS providers when calculating the net present value (NPV) of their project (Propfe et al., 2012). Other
aspects include, for example, the psychological factors associated with operating an EV (Kim et al., 2015). Finally, national differ-
ences must be taken into consideration. Similar to the varying market penetration of vehicle sharing or e-mobility services in different
regions today, the growing EVCS market could be supported or hindered by various governmental requirements. Actual and forecast
fuel and electricity costs, general transportation price levels as well as subsidies and legislative changes play important roles. Besides
the costs for the vehicle itself, the cost for setting up an adequate charging infrastructure has to be taken into consideration (He et al.,
2017). This also falls within municipal responsibility as CS enterprises may use free public charging infrastructures instead of in-
stalling their own.

4. Modeling EVCS

Following the recent publication of discrete event simulation modeling in this research area (Fassi et al., 2012; Burmeister et al.,
2015; Sebastiani et al., 2014; Clemente et al., 2013), our approach was to create a solid simulation tool to address and overcome some
of the issues presented in Sections 2 and 3.2. First of all, the recharge up to 100% SOC between every trip should be avoided. Our
model roughly followed the approach of Clemente et al. (2013), but with a different scope. Instead of surveying the β service level
solely as the major key performance indicator of an entire CS network, we focused on a more detailed vehicle charge routine, the type
of vehicle, and the impact on profitability in different market environments. Our study is independent of the network size of any case
study, as it was framed on a generic base and draws on the data from a single vehicle in its cost analysis. The queue based discrete
event simulation software used was Rockwell Automation Arena (version 15.00, academic license/academic lab package). This type
offered the best possible precision by simulating every event while maintaining low computational effort. In our model phases with
few events for example at night saw significant acceleration. Another noteworthy advantage of the utilized software was the com-
bination of stochastic and deterministic simulation. This means identical simulation runs always provided identical results. Changes
to the model were directly visible without stochastic interference. Still multiple replications within one simulation run were based on
pseudo random values generated by the software from any desired random distribution. A thorough literature review on various
simulation types and their advantages can be found in Jahangirian et al. (2010).
The core of the presented model was a realistically implemented charging routine. The model as a whole was created within one
unified model window, but divided into two threads (see Fig. 1). The first sub-model contained the actual CS simulation with specific
stochastic demand, an availability-check routine, and trip generation of stochastic length. The customers were represented as entities
and the vehicles as resources. The second model thread was responsible for adjusting the SOC of vehicles based on the actual vehicle
state. For example, driving led to a decrease in the battery energy level. Idling after return while connected to a free charger increased
the SOC. Simulation subroutines determined a maximum SOC of 100% and, more importantly, a minimum level of 30%. This
assumed threshold represents the limit to allow for typical CS trip length with the mini vehicle class and is a psychological important
limit for users as well. Whenever a vehicle fell below that threshold during operation, it was forced to recharge up to a certain SOC
(here 50%) before it could reenter the pool of available vehicles. In the second model thread, entities were generated with a fixed
time interval in order to facilitate the energy-control process. This thread was replicated frequently and independently for every

379
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Fig. 1. Simulation flow diagram.

vehicle in the simulation. Manually setting up this subroutine was necessary as the transportation modules available did not allow for
sophisticated battery charge simulation.
In contrast to conventional vehicles, the technical limitations of EVs noted above were crucial for the simulation model. In order
to offer acceptable real-world applicability, the simulation was conducted with different settings for different types of vehicle. Each
setting represented utilization of a homogenous fleet consisting of only one vehicle class. The results were then processed according
to the situations in three different markets. We used three types of EV classes – namely ‘Mini’, ‘Small’ and ‘Medium’ – as well as one
gasoline car class for comparison. These were fully generic vehicles modeled on the basis of real-world examples. This enabled us to
avoid incompatibilities due to e.g. various national driving cycles, technical variations within the same vehicle type, or the use of
completely different brands and models as well as different models being launched on different markets. However, classifying EVs is a
rather difficult task. Typical measures like the size or price of a car are not necessarily related to its driving range. Technical progress
has been constantly increasing driving ranges, so it seems to be only a matter of time until new vehicles become available that
combine the lower prices of current entry-level models with the longer ranges of more advanced cars. Our examples presented three
cars with the individual specifications given in Table 1. Other features are assumed to be equal across all vehicles. This assumption
requires further attention in real world applications. For example, the smallest class may not be capable of carrying the same number
of passengers or luggage and thus, offering not the same usability. Even though our vehicles were based on real-world examples,
range and energy consumption did not follow official manufacturer data. Instead, worst-case scenarios were considered that reflected
inefficient inner-city driving and operation in winter conditions. The recharge time and energy consumption of EVs may not run
parallel to each other under real-world conditions. The required time for charging a certain energy amount may vary: for example,
charging from 50% to 55% may take less time than from 95% to 100%. We assumed a linear behavior across the range below 80%
SOC and reduced this recharge rate by 50% above this threshold. This assumption emulates the real world effect observed with any
EV (Affanni et al., 2005). In order to perform a comparison of the three EV types with a generic gasoline-fueled alternative, the mini-
class base price and a fuel consumption of 7L/100Km was assumed in the calculations. Similarly to EVs, this represented a worst-case
scenario including inefficient inner-city driving and cold temperatures. Therefore the maximum achievable EV range according to the

Table 1
Three vehicle types for simulation.
Type Range Energy consumption Charge time 30–100% Base price in $000’s plus VAT in different regions

km 70% km kWh/100 km h A B C

‘Mini’ 75 52 14 2.5 20 23 10
‘Small’ 130 91 16 3 27 30 17
‘Medium’ 240 168 20 5 33 36 24

380
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Fig. 2. Literature review on CS demand patterns.

manufacturer data where taken and reduced by 30% (Yuksel and Michalek, 2015). Depending on the actual temperatures, vehicle
type and battery technology this effect may be more or less significant in real world scenarios.
Our proposed simulation was not based on a real world case study, even though the generic demand must align with real world
observations as close as possible (Ciari et al., 2012). At first it was based on literature findings presented in Fig. 2.
The demand patterns observed by A: He et al. (2017) B: Li et al. (2016) C: Schmöller et al. (2015) D: Schreier et al. (2015) E: civity
Management Consultants GmbH & Co. KG (2014) and F: Nair and Miller-Hooks (2010) all represent similar patterns. Unfortunately,
demand data from studies found was difficult to apply into our simulation as the available trip data were not linked to definite
vehicles or only average numbers were present. Therefore, we investigated online reservation systems of free-floating and station-
based CS networks in Germany. This observation was transformed into the scheduled base load demand presented in Fig. 3. The
observed demand fluctuations were realized through two additional stochastic components adding random demand. In total the
simulation model consists of one create module with scheduled demand and two more based on uniform distributions between 30
and 120 min and 120 and 480 min.
The generation system described represents the demand for one station with three vehicles where we also had the focus during
our real world observations. This avoided uncertainties during up or downscaling the actual demand. The final demand simulated by
the software is presented in Fig. 4. It has to be noted due to trip times with less than one hour a single vehicle served multiple
customers per hour. If capacity is exhausted, it was assumed 30% of the customers try again to perform their trip 30 min later while
others leave.
Furthermore we aimed to maintain a desired vehicle trip station ratio (VTSR), as introduced by Kek, Alvina G. H. et al. (2009).
This performance indicator is calculated from the total number of vehicles (A), the total number of trips per day (B) and the number
of stations (C) through the relation A/(B × C) (Kek, Alvina G. H. et al., 2009). Using this key performance indicator as a modeling
parameter made network size less relevant, as initially invented by Barth and Todd (1999). In this model, a VTSR of 0.15 was the
target. The relatively simple model structure increased reproducibility. Trip times were taken from Clemente et al. (2013) in ac-
cordance with the findings of Kim et al. (2017). It was based on a triangular distribution (min, med, and max) with the parameters
(13, 20, and 30) minutes, respectively. Trip distance was assumed to be linearly correlated to trip time, thus determining electricity or
fuel consumption. As typical CS characteristics, relatively short distances per trip are analyzed by Costain et al. (2012). Finally, the
run-time settings were determined. Warm-up time did not have a noticeable influence on the proposed model, and was therefore not
required. Nocturnal idle phases led to fully loaded SOC levels every morning, thus offering identical starting conditions each day. The
simulation run-time was determined as a multiple of one week in order to cover the different demand situations on both weekends
and weekdays. In terms of the results, a long run-time of multiple weeks acted in a similar fashion to additional replications. A higher

Fig. 3. Arena software base load demand schedule.

381
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Fig. 4. Boxplot of the simulated demand.

number of replications negated the impact of unusual stochastic anomalies. Short processing times of far less than 1 min for a single
run on an average desktop computer allowed for a high number of iterations. With 10 replications and a run-time of 4 weeks, the
maximum variation of the results lay below 5% for every experimental case, thus maintaining a 95% confidence interval.
After setting up all the simulation routines, the verification and validation process was initiated. Best practice procedures are
carried out according to Kleijnen (1995). The alignment between the inputted real-world demand data and the simulated demand
was checked with the following linear regression analysis (see Table 2). The observed real-world average vehicle stock per hour
represents the independent variable and the generated demand by our simulation model is the dependent variable.
The residuals show no patterns and are normally distributed (µ= 0.000; σ = 0.322) with a square error of 0.005. According to the
regression results the simulated demand in our model represented the real world demand very well. Model creation was initiated on
the basis of a minimalistic but fully functional simulation model. Further extensions were added and the credibility of new results was
checked frequently. The effect of predictable exceptions and the extreme values of certain parameters were also monitored. This
included, for example, a zero kilometer vehicle range or the zero-level availability of charging stations. Furthermore, we utilized the
internal visualization tools, error-check routines and the report function provided by the Arena software framework. Results of the
simulation were compared with results from Clemente et al., (2013) and indicated identical trends, e.g. regarding the service level.
All of the primary simulation results obtained were processed with respect to their impact on the NPV. Calculations were made for
three different markets. The market data assumed in Table 3 corresponded approximately to the current situations in North America,
e.g. the United States (A), Western Europe, e.g. Germany (B), and East Asia, e.g. China (C). The average inflation rate forecast on all
markets for electricity and gasoline prices was assumed to be 1% per year (Plötz et al., 2015).
The working life of one vehicle was set at 4 years and the battery lifecycle was assumed to be equal to the vehicle working life. The
resale value after 4 years was rather difficult to predict for EVs, as it depends on mileage and the battery condition (Mahmoudzadeh
Andwari et al., 2017). Due to contradictory research findings, we assumed a resale value equal to 40% of the original price, which is
comparable with conventional cars (Propfe et al., 2012). The impact of all assumptions is further reviewed in the sensitivity analysis.

5. Experiments and results

The simulation results obtained in all experiments verified the expected importance of vehicle range. Fig. 5 shows a working day
for each of the vehicle types analyzed. Specifically, the mini class EV fell below its 30% SOC limitation twice during the afternoon.
The short recharge times during the idle phases were not enough to compensate the energy consumption during the active times of
operation. As a result, the vehicle was forced to recharge up to 50%, and was unavailable for customers during that time in the
evening. The small vehicle type also encountered the same situation, but only once and at the end of the day. Only a small proportion
of demand was lost, as the demand decreased in the evenings. The medium class vehicle offered enough range to stay available during
the whole day. This meant that it exhibited availability equal to that of conventionally fueled vehicles.

Table 2
Regression analysis for verification.
Regression accuracy ANOVA significance Coefficient reliability

R2
0.873 F(1,22)= p = 0.000 t = −12.283 p = 0.000
Adjusted R2 0.867 150.876 Y-Intercept reliability

Std. error 0.337 t = 21.725 p = 0.000

382
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Table 3
Three sample markets (OECD, 2018).
Market Electricity price Gasoline price Discount rate Price level indices
$/kWh $/liter % %

A 0.21 0.75 2 113


B 0.33 1.6 1 98
C 0.10 1.1 3 58

Fig. 5. Simulation results.

In order to facilitate further insight, three additional vehicle range set-ups were tested and added to the chart in Fig. 6. It became
apparent that in our simulation, a range availability of 130 km was sufficient to satisfy the same proportion of demand as did
conventionally fueled cars. This range equaled a theoretical total range of 186 km due to the 30% minimum SOC constraint. Below
that range, the achievable β service level drops. In our context the β service level is defined as the ratio between satisfied demand and
total demand. The logarithmic trend line illustrates similar outcomes to Clemente et al. (2013). The relatively low but realistic degree
of utilization arose from low demand phases at night. The unsatisfied demand proportion occurred during random demand spikes.
Our demand generation process promoted randomly occurring but also realistically short inter-arrival times. In these situations, the
number of vehicles was too low – no matter what range was theoretically available.
Furthermore, Fig. 6 illustrates the approach to hitting a target β service level equal to that of the medium-type vehicle option.
More vehicles of the mini and small types would have been necessary to compensate for the drop in the service level. These additional
vehicles would have increased total fleet investments. Based on the simulation results and the assumed market data, the NPV was
calculated for all alternatives (see Fig. 8). The significantly diverse price level indices (see Table 3) posed a noticeable impact. For
clarity, Fig. 7 provides the necessary yearly income required to hit the break-even point in the NPV calculation.
Within the EV class, the mini vehicle scored the best results – even though it offered the worst performance in the CS simulation.
The reason for this observation lay in the cost structure assumed. This behavior can also be found in real-world vehicle pricing. In
other words, it seemed more profitable to operate a larger CS fleet of short-range mini cars than a smaller fleet consisting of vehicles
with longer ranges. The other two models were purchased at disproportionately high costs. This conclusion is based on the as-
sumption that a given demand should be satisfied and a desired service level has to be reached which can be achieved by operating a
higher number of small ranged cars or a smaller number of cars with a longer range equally. According to our NPV calculation the
first option generates more profit for the CS company. Furthermore, there is higher flexibility when assigning vehicles to stations. A
disadvantage of this approach is the higher space requirement which can be a problem in city centers. Especially in one-way systems

Fig. 6. Average β service level and utilization chart with logarithmic trend lines.

383
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Fig. 7. Required break-even revenue per vehicle and year for NPV = 0.

vehicles may compete against each other for chargers so advanced parking reservation policies will be required (Kaspi et al., 2014).
The sensitivity analysis in the next section also explains this issue in detail. Furthermore, EVs appeared to be superior choices when
compared to the conventionally fueled vehicles in markets B and C. The situation in market A was determined (most notably) by
relatively low gasoline prices (see Table 3).
If a fixed yearly profit is set and scaled according to the national price plateau also presented in Table 3, Fig. 8 would suggest
investments in EVs for all markets. The revenue achieved per car is the most important factor determining the NPV. Small adjust-
ments in the usage fee can enhance profitability considerably. Real-world examples show that many CS providers rely on manifold
pricing models based on travel time and distance, in association with monthly basic fees. However, usage fees are largely determined
by competitors and mobility alternatives like public transport or private car ownership. It is questionable whether customers will
accept higher usage fees if electric vehicles are operated (Xu et al., 2017; Yoon et al., 2017).
Any costs incurred for the required charger infrastructure were not considered. However, a higher NPV means higher financial
resources to cover those investments. In other words the advantageous revenue gap between EVs and conventional cars on markets B
and C from Fig. 7 may be invested into charger infrastructure. On market C this gap amounted to $1677.59 per operated vehicle and
year for vehicle type ‘mini’ and $1086.50 for the ‘small’ vehicle type or $979.85 and $661.69 on market B respectively.

6. Discussion

Of course, the assumptions made influence the results obtained. In order to identify the primary drivers of the outcomes pre-
sented, a sensitivity analysis was conducted (see Fig. 9). There were three groups of parameters in our study. Firstly, three customer
related values - arrival times, trip length and retry rate. This group was summarized into utilization for the reason that they all lead to
higher/lower vehicle utilization within the limits of available vehicle capacity. If a target service level is prescribed, either more or
less vehicles must be operated. This could increase or decrease the investment cost dramatically. However, as shown above, more
units of affordable low-range cars provide better results than fewer long-range vehicles. If demand surpasses vehicle capacities
additional vehicles would be required and the whole model scales up. The NPV was calculated per vehicle so the number of vehicles
was irrelevant as long as economies of scale were not considered. The second parameter group included vehicle specifications -
recharge rate, vehicle range (consisting of energy consumption and battery capacity) and minimum required SOC. The recharge rate
showed only minor impact, as most of the recharge process, took place at night. The minimum required SOC is left out, as it could
barely be adjusted in real-world applications. Lower values increased the risk of not being able to perform a desired trip while higher

Fig. 8. Net present value.

384
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Fig. 9. Sensitivity analysis.

values are ineffectual. The vehicle range instead was very important as it determines the trip capacity a car provided and was
therefore included into the sensitivity analysis. Thirdly, all parts of the cost model, such as prices and discount rate are calculated on
top of the obtained simulation results. To this end, all parameters were varied within the same relative range and the percentage
effect on the NPV was calculated. The discount rate and the rate of inflation had practically no noticeable effect. The effect of
electricity cost is almost as high as the one of utilization. Still the energy consumption of EVs is far lower than the relative fuel
consumption of conventionally fueled vehicles, and electricity is traded at a lower price on all markets (per unit of energy). De-
preciation was a decisive factor for the resale value of cars, and therefore had a significant effect like of the vehicle price itself, though
to a lesser degree. In real-world applications, government subsidies or manufacturer promotions can easily shift the investment into a
relatively profitable zone for CS enterprises. Still the most important factors were vehicle price and vehicle range. Especially re-
duction of the range leads to pool model performance by reducing the β service level. Compensating this effect demands additional
vehicles, thus raising costs drastically.
The proposed discrete event simulation model generates reasonable results for an EVCS cost analysis. As expected, the low range
of entry level vehicles becomes an issue if, due to high utilization, the vehicle has no time to recharge and maintain a certain SOC
level. However, lowering utilization levels during demand spikes by providing additional vehicles avoids this situation. CS operators
are thus encouraged to transform existing CS fleets of conventionally fueled cars into EV fleets in a gradual manner. In this way,
gasoline-fueled cars can still be deployed if necessary. Furthermore, the battery capacities of many of the EV types currently on the
market are sufficient for operation in CS networks without any drawbacks. Future technical progress will provide vehicles that are
even more suitable, while widespread adoption of EVs will likely bring down purchase prices. Those are still the major issue de-
termining further EV market penetration (Yoon and Cherry, 2018). Unfortunately, the smallest vehicle class might not offer the same
characteristics as larger cars regarding flexibility and usability. A smaller number of seats and smaller luggage compartment could
prevent the vehicle to take over trips a larger car could handle. This comes also into effect if vehicle sharing is combined with ride
sharing (Barth et al., 2004). EV charger infrastructure was not considered in the calculations presented, as the focus of this paper is on
investments in vehicles only. To include EV chargers would increase complexity significantly. Firstly, the charger infrastructure has a
longer operating life than the vehicles. The number of chargers depends on the network size and number of vehicles. Moreover,
multiple vehicles can share one charger with multiple plugs, making the assignment of costs more difficult. A key performance
indicator would be required that would set the number of stations, vehicles and trips in relation to the number of chargers required,
similar to the relative VTSR measurement. This would also depend on real-world applications because municipal/public charger
infrastructure could be used if available. Finally, there may be incompatibilities with respect to plug and voltage standards between
different vehicle makes. However, our cost calculation at least identifies the surplus that may be invested in chargers. Further
research is necessary in this field, but initial insights indicate that the number of available chargers is not of paramount importance
for general vehicle performance unless it is particularly low. The reason is that the idle phases at night are long enough to permit
complete charging for multiple vehicles. Additional downtimes for planned maintenance or sudden accidents are also disregarded in
the model. Such cases would further reduce availability in real-world applications.

7. Conclusion

The applicability of electric vehicles in car sharing systems has been proven. Under certain assumptions, state-of-the-art electric
vehicle sharing networks are ready to compete with conventionally fueled cars. It has been shown that profitability is primarily
dependent on the vehicle price, range, and potentially achievable usage fees. Our research findings can be summarized in the
following practical implications:

• Current technical vehicle range limits may be resolved through technological progress, or could be avoided by CS fleets utilizing a
larger number of vehicles with shorter ranges without financial drawbacks.
• Cost analysis encourages the operation of fleets based on less expensive vehicles with shorter ranges, even though this means that
385
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

the fleet size must be increased. However, superior vehicles may become more affordable through subsidies or promotions in the
future. Furthermore, penalty taxes on the purchase or operation of conventional vehicles in urban areas may also increase the
attractiveness of EVs.
• Some markets offer better conditions than others for operating EVs in vehicle sharing networks. Our cost analysis indicates that for
markets in East Asia and Europe, EVs may be recommended strongly, while North America still offers good conditions for the
operation of gasoline-fueled cars. However, from an operator’s viewpoint, other market indicators point to a noticeable general
financial advantage for vehicle sharing networks also in North America.
• Widespread charger infrastructures ready to serve car-sharing fleets have not yet been implemented in many cities. CS companies
need long-term technical standards that they can rely on to allow them to perform efficient location planning when setting up
charger infrastructures – as described by Brandstätter et al. (2017).
• Legislation is encouraged to promote electric vehicle sharing in order to fulfill politically desirable EV market shares, for example
by supporting the development of city-wide charger infrastructures that would also be open to vehicles of CS companies.

The findings focus on CS vehicles and do not take the costs associated with charger infrastructure into consideration, as they are
difficult to allocate to individual cars. Overall, our paper contributes to the process of transforming current transportation systems
into sustainable future transportation modes, ready to solve issues arising from ongoing urbanization, population growth and the
need for more environmentally friendly mobility solutions. Future research can be based on the presented findings which offer
manifold opportunities for model extensions. For example, more realistic modeling of charger infrastructures and vehicle relocations
for electric one-way CS networks would allow for better practical application. It might be fruitful to concentrate on one market and
vehicle option only, and/or to investigate other factors such as battery degeneration over time, time-dependent electricity costs, and
dynamic pricing models for CS usage. If other parameters e.g. vehicle type are fixed, the model size can be increased to realistic
numbers in order to survey economies of scale of larger EVCS networks. Finally, an electric vehicle sharing life-cycle assessment –
which can only be found for electric mobility in Helmers et al. (2017) – could offer interesting insights and support future electric
vehicle sharing.

Acknowledgements

The authors would like to thank anonymous reviewers for their helpful contribution that greatly improved the paper. This
research did not receive any specific grant funding from agencies in the public, commercial, or not-for-profit sectors.

References

Abdelkafi, N., Makhotin, S., Posselt, T., 2013. Business model innovations for electric mobility – what can be learned from existing business model patterns? Int. J.
Innovation Manage. 17 (1) 1340003-1-1340003-41.
Affanni, A., Bellini, A., Franceschini, G., Guglielmi, P., Tassoni, C., 2005. Battery choice and management for new-generation electric vehicles. Inst. Electr. Electron.
Eng. - Trans. Ind. Electron. 52 (5), 1343–1349. http://dx.doi.org/10.1109/TIE.2005.855664.
Barth, M., Todd, M., 1999. Simulation model performance analysis of a multiple station shared vehicle system. Transport. Res. Part C: Emerging Technol. 7 (4),
237–259. http://dx.doi.org/10.1016/S0968-090X(99)00021-2.
Barth, M., Todd, M., Xue, L., 2004. User-based vehicle relocation techniques for multiple-station shared-use vehicle systems. TRB 2004 Annual Meeting.
Boyacı, B., Zografos, K.G., Geroliminis, N., 2015. An optimization framework for the development of efficient one-way car-sharing systems. Eur. J. Oper. Res. 240 (3),
718–733. http://dx.doi.org/10.1016/j.ejor.2014.07.020.
Brandstätter, G., Kahr, M., Leitner, M., 2017. Determining optimal locations for charging stations of electric car-sharing systems under stochastic demand. Transport.
Res. Part B: Methodol. 104, 17–35. http://dx.doi.org/10.1016/j.trb.2017.06.009.
Burmeister, F., Schnieder, L., Kurczveil, T., 2015. Simulation based studies on the integration of battery-electric vehicles in regional bus services. Intell. Transport. Syst.
- Inst. Electr. Electron. Eng. 2015. http://dx.doi.org/10.1109/ITSC.2015.219.
Cepolina, E.M., Farina, A., 2012. A new shared vehicle system for urban areas. Transport. Res. Part C: Emerging Technol. 21 (1), 230–243. http://dx.doi.org/10.1016/
j.trc.2011.10.005.
Ciari, F., Schuessler, N., Axhausen, K.W., 2012. Estimation of carsharing demand using an activity-based microsimulation approach: model discussion and some
results. Int. J. Sustain. Transport. 7 (1), 70–84. http://dx.doi.org/10.1080/15568318.2012.660113.
civity Management Consultants GmbH & Co. KG, 2014. Urbane Mobilität im Umbruch?: matters No. 1 Verkehrliche und ökonomische Bedeutung des Free-Floating-
Carsharing, 46p.
Clemente, M., Fanti, M.P., Iacobellis, G., Ukovich, W., 2013. Modeling and simulation of an electric car sharing system. In: 25th European Modeling and Simulation
Symposium, pp. 587–592. http://www.msc-les.org/proceedings/emss/2013/EMSS2013_587.pdf.
Costain, C., Ardron, C., Habib, K.N., 2012. Synopsis of users’ behaviour of a carsharing program: a case study in Toronto. Transport. Res. Part A: Policy Pract. 46 (3),
421–434. http://dx.doi.org/10.1016/j.tra.2011.11.005.
Engel-Yan, J., Passmore, D., 2013. Carsharing and car ownership at the building scale. J. Am. Plan. Assoc. 79 (1), 82–91. http://dx.doi.org/10.1080/01944363.2013.
790588.
Fairley, P., 2013. Car sharing could be the EV's killer app. Inst. Electr. Electron. Eng. - Spectrum 50 (9), 14–15. http://dx.doi.org/10.1109/MSPEC.2013.6587173.
Farid, A.M., 2017. A hybrid dynamic system model for multimodal transportation electrification. Inst. Electr. Electron. Eng. - Trans. Control Syst. Technol. 25 (3),
940–951. http://dx.doi.org/10.1109/TCST.2016.2579602.
Fassi, A.E., Awasthi, A., Viviani, M., 2012. Evaluation of carsharing network’s growth strategies through discrete event simulation. Expert Syst. Appl. 39 (8),
6692–6705. http://dx.doi.org/10.1016/j.eswa.2011.11.071.
Fuentes, M., Fraile-Ardanuy, J., Risco-Martín, J.L., Moya, J.M., 2017. Feasibility study of a building-integrated PV manager to power a last-mile electric vehicle sharing
system. Int. J. Photoenergy 2017 (25), 1–12. http://dx.doi.org/10.1155/2017/8679183.
Gavalas, D., Konstantopoulos, C., Pantziou, G., 2015. Design and management of vehicle sharing systems: a survey of algorithmic approaches. In: Obaidat, M., Nicop
olitidis, P. (Eds.), Smart Cities and Homes: Key Enabling Technologies. Elsevier Science.
He, L., Mak, H.-Y., Rong, Y., Shen, Z.-J.M., 2017. Service region design for urban electric vehicle sharing systems. Manuf. Service Oper. Manage. 19 (2), 309–327.
http://dx.doi.org/10.1287/msom.2016.0611.
Helmers, E., Dietz, J., Hartard, S., 2017. Electric car life cycle assessment based on real-world mileage and the electric conversion scenario. Int. J. Life Cycle

386
S. Illgen, M. Höck Transportation Research Part D 63 (2018) 377–387

Assessment 22 (1), 15–30. http://dx.doi.org/10.1007/s11367-015-0934-3.


Jahangirian, M., Eldabi, T., Naseer, A., Stergioulas, L.K., Young, T., 2010. Simulation in manufacturing and business: a review. Eur. J. Oper. Res. 203 (1), 1–13. http://
dx.doi.org/10.1016/j.ejor.2009.06.004.
Kaspi, M., Raviv, T., Tzur, M., 2014. Parking reservation policies in one-way vehicle sharing systems. Transport. Res. Part B: Methodol. 62, 35–50. http://dx.doi.org/
10.1016/j.trb.2014.01.006.
Kek, Alvina G.H., Cheu, R.L., Meng, Q., Fung, C.H., 2009. A decision support system for vehicle relocation operations in carsharing systems. Transp. Res. Part E:
Logistics Transp. Rev. 45 (1), 149–158. http://dx.doi.org/10.1016/j.tre.2008.02.008.
Kihm, A., Trommer, S., 2014. The new car market for electric vehicles and the potential for fuel substitution. Energy Policy 73, 147–157. http://dx.doi.org/10.1016/j.
enpol.2014.05.021.
Kim, D., Ko, J., Park, Y., 2015. Factors affecting electric vehicle sharing program participants’ attitudes about car ownership and program participation. Transp. Res.
Part D: Transp. Environ. 36, 96–106. http://dx.doi.org/10.1016/j.trd.2015.02.009.
Kim, J., Rasouli, S., Timmermans, H.J., 2017. The effects of activity-travel context and individual attitudes on car-sharing decisions under travel time uncertainty: a
hybrid choice modeling approach. Transp. Res. Part D: Transp. Environ. 56, 189–202. http://dx.doi.org/10.1016/j.trd.2017.07.022.
Kleijnen, J.P., 1995. Verification and validation of simulation models. Eur. J. Oper. Res 82 (1), 145–162. http://dx.doi.org/10.1016/0377-2217(94)00016-6.
Kopp, J., Gerike, R., Axhausen, K.W., 2015. Do sharing people behave differently?: An empirical evaluation of the distinctive mobility patterns of free-floating car-
sharing members. Transportation 42 (3), 449–469. http://dx.doi.org/10.1007/s11116-015-9606-1.
Le Vine, S., Lee-Gosselin, M., Sivakumar, A., Polak, J., 2014. A new approach to predict the market and impacts of round-trip and point-to-point carsharing systems:
case study of London. Transp. Res. Part D: Transp. Environ. 32, 218–229. http://dx.doi.org/10.1016/j.trd.2014.07.005.
Li, X., Ma, J., Cui, J., Ghiasi, A., Zhou, F., 2016. Design framework of large-scale one-way electric vehicle sharing systems: a continuum approximation model.
Transport. Res. Part B: Methodol. 88, 21–45. http://dx.doi.org/10.1016/j.trb.2016.01.014.
Lindloff, K., Pieper, N., Bandelow, N.C., Woisetschläger, D.M., 2014. Drivers of carsharing diffusion in Germany: an actor-centred approach. Int. J. Automot. Technol.
Manage. 14 (3/4), 217–245. http://dx.doi.org/10.1504/IJATM.2014.065291.
Luè, A., Colorni, A., Nocerino, R., Paruscio, V., 2012. Green move: an innovative electric vehicle-sharing system. Proc. - Soc. Behav. Sci. - Transp. Res. Arena 48,
2978–2987. http://dx.doi.org/10.1016/j.sbspro.2012.06.1265.
Mahmoudzadeh Andwari, A., Pesiridis, A., Rajoo, S., Martinez-Botas, R., Esfahanian, V., 2017. A review of Battery Electric Vehicle technology and readiness levels.
Renew. Sustain. Energy Rev. 78, 414–430. http://dx.doi.org/10.1016/j.rser.2017.03.138.
Millard-Ball, A., 2005. Car-Sharing: Where and How It Succeeds: 108 Transit Cooperative Research Program. Transportation Research Board. http://dx.doi.org/10.
17226/13559.
Moradi, A., Vagnoni, E., 2018. A multi-level perspective analysis of urban mobility system dynamics: what are the future transition pathways? Technol. Forecast. Soc.
Chang. 126, 231–243. http://dx.doi.org/10.1016/j.techfore.2017.09.002.
Nair, R., Miller-Hooks, E., 2010. Fleet management for vehicle sharing operations. Transport. Sci. 45 (4), 524–540. http://dx.doi.org/10.1287/trsc.1100.0347.
Organisation for Economic Co-operation and Development, 2018. Price level indices (indicator). doi: 10.1787/c0266784-en. Accessed 10 January 2018.
Parzinger, G., Rid, W., Müller, U., Grausam, M. (Eds.), 2016. Elektromobilität im Carsharing: Status Quo, Potenziale und Erfolgsfaktoren, second ed. Nationale
Organisation Wasserstoff- und Brennstoffzellentechnologie, Berlin, pp. 87.
Pelletier, S., Jabali, O., Laporte, G., Veneroni, M., 2017. Battery degradation and behaviour for electric vehicles: review and numerical analyses of several models.
Transport. Res. Part B: Methodol. 103, 158–187. http://dx.doi.org/10.1016/j.trb.2017.01.020.
Plötz, P., Gnann, T., Wietschel, M., Ullrich, S., 2015. How to foster electric vehicle market penetration? A model based assessment of policy measures and external
factors. European Council for an Energy-Efficient Economy, Paris, pp. 843–853.
Propfe, B., Redelbach, M., Santini, D.J., Friedrich, H., 2012. Cost analysis of plug-in hybrid electric vehicles including maintenance & repair costs and resale values.
International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium 26.
Reining, C., Loose, W., Allebrod, A., 2014. Elektrofahrzeuge in CarSharing-Flotten - Chance und Herausforderung. Eine Idee setzt sich durch! 25 Jahre CarSharing,
97–109.
Schmöller, S., Weikl, S., Müller, J., Bogenberger, K., 2015. Empirical analysis of free-floating carsharing usage: the Munich and Berlin case. Transport. Res. Part C:
Emerging Technol. 56, 34–51. http://dx.doi.org/10.1016/j.trc.2015.03.008.
Schreier, H., Becker, U., Heller, J., 2015. Evaluation CarSharing (EVA-CS), Berlin.
Sebastiani, M.T., Luders, R., Fonseca, K.V.O., 2014. Allocation of charging stations in an electric vehicle network using simulation optimization. In: Proceedings of the
2014 Winter Simulation Conference, pp. 1073–1083. http://doi.org/10.1109/WSC.2014.7019966.
Shaheen, S.A., Chan, N.D., Micheaux, H., 2015. One-way carsharing’s evolution and operator perspectives from the Americas. Transportation 42 (3), 519–536. http://
dx.doi.org/10.1007/s11116-015-9607-0.
Shaheen, S.A., Cohen, A.P., 2013. Carsharing and personal vehicle services: worldwide market developments and emerging trends. Int. J. Sustain. Transport. 7 (1),
5–34. http://dx.doi.org/10.1080/15568318.2012.660103.
Vervaeke, M., Calabrese, G., 2015. Prospective design in the automotive sector and the trajectory of the Bluecar project: an electric car sharing system. Int. J. Veh. Des.
68 (4), 245. http://dx.doi.org/10.1504/IJVD.2015.071083.
Weikl, S., Bogenberger, K., 2015. A practice-ready relocation model for free-floating carsharing systems with electric vehicles – mesoscopic approach and field trial
results. Transport. Res. Part C: Emerging Technol. 57, 206–223. http://dx.doi.org/10.1016/j.trc.2015.06.024.
Wielinski, G., Trépanier, M., Morency, C., 2016. Electric and hybrid car use in a free-floating carsharing system. Int. J. Sustain. Transport. 11 (3), 161–169. http://dx.
doi.org/10.1080/15568318.2016.1220653.
Xu, F., Liu, J., Lin, S., Yuan, J., 2017. A VIKOR-based approach for assessing the service performance of electric vehicle sharing programs: a case study in Beijing. J.
Cleaner Prod. 148, 254–267. http://dx.doi.org/10.1016/j.jclepro.2017.01.162.
Yoon, T., Cherry, C.R., 2018. Migrating towards using electric vehicles in campus-proposed methods for fleet optimization. Sustainability 10 (2). http://dx.doi.org/10.
3390/su10020285.
Yoon, T., Cherry, C.R., Jones, L.R., 2017. One-way and round-trip carsharing: a stated preference experiment in Beijing. Transp. Res. Part D: Transp. Environ. 53,
102–114. http://dx.doi.org/10.1016/j.trd.2017.04.009.
Yuksel, T., Michalek, J.J., 2015. Effects of regional temperature on electric vehicle efficiency, range, and emissions in the United States. Environ. Sci. Technol. 49 (6),
3974–3980. http://dx.doi.org/10.1021/es505621s.
Zhang, G., Ge, S., Yang, X.-G., Leng, Y., Marple, D., Wang, C.-Y., 2017. Rapid restoration of electric vehicle battery performance while driving at cold temperatures. J.
Power Sources 371, 35–40. http://dx.doi.org/10.1016/j.jpowsour.2017.10.029.

387