- Email: [email protected]
Electric bike sharing: simulation of user demand and system availability
Journal of Cleaner Production xxx (2013) 1e8
Contents lists available at ScienceDirect
Journal of Cleaner Production journal homepage: www.elsevier.com/locate/jclepro
Electric bike sharing: simulation of user demand and system availability Shuguang Ji a, Christopher R. Cherry b, *, Lee D. Han b, David A. Jordan b a b
University of Tennessee, Department of Industrial and Systems Engineering, Knoxville, TN, USA University of Tennessee, Department of Civil and Environmental Engineering, Knoxville, TN, USA
a r t i c l e i n f o
a b s t r a c t
Article history: Received 14 May 2013 Received in revised form 18 September 2013 Accepted 20 September 2013 Available online xxx
This paper describes the operational concepts and system requirements of a fully automated electric bike (e-bike) sharing system demonstrated through a pilot project at the University of Tennessee, Knoxville (UTK) campus (deployed in September 2011). This project is part of a movement to develop a sustainable transportation system, and is one of the green initiatives on UTK campus. E-bikes are more energy efficient and produce fewer greenhouse gas (GHG) emissions per person compared to other transport modes such as car, bus, and motorcycle. Without empirical demand information for an e-bike sharing system, we simulated the operations of such a system to gain insights during the design process before field deployment. The simulation exercise focused on three critical demand parameters e distributions of trip rates, trip lengths, and trip durations e and coupled them with supply parameters e number of ebikes, number of swappable batteries, and battery recharging profiles. The primary purpose of these simulations is to evaluate the efficiency of an off-board battery recharging system, where the depleted battery is removed from an e-bike upon its return and inserted into one of the charging slots at the kiosk. We tested various scenarios with different number of batteries always maintaining an initial condition with the battery to e-bike ratio greater or equal to 1.0 to ensure battery availability. We applied empirical battery recharging rates and system operations rules to determine the number of e-bikes and batteries available under different potential demand situations, with a focus on optimizing the number of batteries to meet user demands. By adjusting input parameters, numerous scenarios were simulated for sensitivity analysis. Based on the results of the simulation, this paper presents a cost constrained e-bike sharing system design that can maintain a high level of system reliability (e-bike and battery availability) through optimal battery charging and distribution management. We found that high demand scenarios require multiple swappable batteries per e-bike to reasonably meet the maximum demand. Trip duration has the most influence on e-bike and battery availability, followed by trip rate, and then trip length. Ó 2013 Elsevier Ltd. All rights reserved.
Keywords: Electric bike sharing System operations Monte Carlo simulation E-bike
1. Introduction Bike sharing is a new provision model of public/private transportation that has emerged in the past several years. As of January 2012, there are 15 automated bike sharing programs in the U.S. involving 5238 bicycles and 172,070 users (Shaheen et al., 2013). Bike sharing is among the most environmentally sound models of transport that provides public transportation services for short trips that are often inadequately served by other modes of fixedroute public transit (Hampshire and Marla, 2012; Shang et al., 2010). Bike sharing systems have been overcoming many operational challenges in the past decade to provide fully automated,
* Corresponding author. Tel.: þ1 865 974 7710; fax: þ1 865 974 2669. E-mail address: [email protected] (C.R. Cherry).
secure, and cost-effective systems and have been deployed in over 100 cities worldwide (Kim et al., 2012). Traditional bicycles in bike sharing systems range in price, with mainstream systems offering bicycles for about US$1000 (Shaheen et al., 2013). Electric bikes (ebikes) range in price as well but typically carry a price premium of approximately 2 for a similarly equipped bicycle. Formal bike sharing systems have existed for nearly half a century, with various levels of success. Few studies have systematically estimated demand or defined operational parameters (Kaltenbrunner et al., 2010). There have been three generations of evolution, driven mostly by advances in technology. The first generation began in Amsterdam in 1965, where stationless bicycles could be borrowed and left anywhere in the city, to be borrowed again by the next individual. It was unsuccessful due to vandalism and theft. The second generation, born in Denmark in 1991, allowed bicycles to be picked up and returned to several central locations
0959-6526/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jclepro.2013.09.024
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
2
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
Fig. 1. Typical pedal e-bike.
with a coin deposit. Theft was also a problem largely due to the anonymity of the user. Third generation bike sharing was born in Portsmouth University in England and involved several technological improvements such as bike racks that locked automatically, on-board electronics, swipe cards, and telecommunication capabilities. In 2005 and 2007 respectively, Lyon and Paris, France launched highly successful third generation bike sharing programs that grew to over 15,000 and 20,000 bicycles respectively. Today, one bike sharing program per month is being created somewhere in the world (DeMaio, 2009; Shaheen et al., 2010). Beginning in 2008, cities outside of Europe began to launch third generation programs. Rio de Janeiro launched a pilot public bike sharing program in 2009. Several others followed in South America and Asia. Some of the largest bike sharing networks are in cities such as Hangzhou (40,000 bicycles and 1700 stations) and Wuhan, China (13,000 bicycles and 516 stations). In North America, Montreal, Denver, Minneapolis, Boston, and Washington D.C. have launched successful third generation bike sharing programs in recent years. Dozens of other North American cities are in planning stages for installing bike sharing systems (Krykewycz et al., 2010; Schroeder et al., 2009). Simultaneously, e-bikes have gained popularity in many regions of the world. Some researchers suggest that shared e-bikes could provide an even higher level of service compared to existing bike
sharing systems, while maintaining low environmental impacts. Based on recent research findings, benefits such as the ability to travel longer distances and over hills with less fatigue and sweat, compared to traditional bicycles, may help overcome some of the barriers to bicycling. In addition, surveys of e-bike riders reveal that most e-bike owners favor e-bikes over traditional bicycles or conventional motorized vehicles (Dill and Rose, 2012). E-bikes are often regulated like bicycles in most jurisdictions. Ebikes, particularly pedal assist e-bikes, look and operate much like traditional bicycles (Fig. 1). Pedal assist e-bikes offer some of the health benefits of bicycling, requiring relatively modest amounts of physical activity (Gojanovic et al., 2011; Louis et al., 2012; Sperlich et al., 2012; Theurel et al., 2012). However, e-bikes are substantially more expensive than similar-quality non-electric bicycles. As such, the e-bike market has not grown as rapidly in the U.S. compared to other countries (Dill and Rose, 2012; Rose, 2012; Weinert et al., 2007a). By spreading the higher cost of e-bikes over a community of many users and over time, e-bike sharing has the potential of overcoming the price barrier as well as the perceived expense of the technology. The provision of e-bikes in a shared environment also introduces electric vehicle technology to potential users without the pressure or commitment of capital purchase. E-bike sharing promotes electric mobility to users of other motorized modes of transport and introduces renewable energy, recharging infrastructure, and safety to the public (Silvester et al., 2013). By affecting mode shifts of users from other motorized modes and because of their light weight, e-bikes are considered a technology towards cleaner transportation (Tonn et al., 2003). E- bikes outperform other motorized transport modes, including bus transit, on energy efficiency and greenhouse gas emission (GHG) rates per passenger kilometer, considering the complete life cycle environmental impacts (Cherry et al., 2009). Moreover, e-bikes can move emissions from tailpipes to power plants that are often away from urban areas, further reducing human exposures to emissions (Ji et al., 2012). Traditional bike sharing systems have been extensively researched in the past few years. Many studies aim to provide policy recommendations to encourage more bicycle trips (Bachand-Marleau et al., 2012; Gleason and Miskimins, 2012; Parkes and Marsden, 2012). Some studies try to employ geographic information system (GIS) to estimate the demand for
Fig. 2. System components of e-bike sharing system.
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
3
Fig. 3. Flow chart of simulation process.
the bike sharing programs (García-Palomares et al., 2012; Hampshire and Marla, 2012; Krykewycz et al., 2010) or develop new models for the configuration of bike sharing systems (Angeloudis et al., 2012; Kim et al., 2012; Lin and Yang, 2011; Maurer, 2012; Nair et al., 2012). Some studies investigate the influences of bicycle infrastructure (e.g., bicycle lanes) on the bike sharing system demand (Buck and Buehler, 2012; Romero et al., 2012). In this paper, we attempt to extend this bike sharing research domain by incorporating e-bikes into the system, focusing
on system configuration, and managing supply to meet demand through a simulation approach. 2. E-bike sharing system description 2.1. System components In September 2011, the University of Tennessee-Knoxville (UTK) installed North America’s first e-bike sharing station as a technical
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
4
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
and operational research pilot test (Langford et al., 2013). This pilot test platform has two stand-alone stations with ten bicycles each and off-board battery charging capability for battery swapping. Each station has the capacity to charge and distribute up to 15 batteries at a time. The e-bike sharing system consists of pedal assist e-bikes, powered by lithium ion (Li-ion) batteries (Fig. 1), and a vending and charging station (Fig. 2). Each pedal assist e-bike has a sensor measuring the rider’s pedaling effort continuously. This information is used by an on-board control device to supplement the rider’s effort with electromechanical power when necessary to help the rider overcome steep hills and long grades without eliminating the need for the rider to pedal. Increased rider effort increases the electric-assist range and reduces fatigue barriers while still providing some health benefits to the rider and maintaining the legal and perceived identity of e-bike as a bicycle. These benefits make the system and vehicles more attractive to casual riders, who may not otherwise consider traditional bicycles as a viable transport mode. Li-ion batteries are the most common in North American and European e-bikes, primarily because of a relatively low weight and high energy density. To extend battery life, maximize stored energy, and minimize the risk of overcharging and overheating, it is advisable for an e-bike sharing system to adopt a slower battery charging system, which also reduces the availability of batteries. Physical security, vending, and charging are the three main functions of the station, which is equipped with an electromechanical locking system that simplifies the check-in/check-out process and verifies when an e-bike is returned and properly secured. The e-bike sharing station provides access to registered users only. Users are required to remove and place the used battery for recharging as a part of the check-in procedure. While this extra step adds user burden, it maximizes recharging activities automatically, independent of the availability of the e-bike.
Table 1 Input variables and statistical distributions for Monte Carlo simulation. Variable
Base-case value
Statistical distribution model
Units
Trip rate Trip length Riding speedc Activity duration Energy consumption rated
20 3.5 9 2 16
Poisson(l)a Beta(2,6)b Normal(9,4) Normal(2,2) Uniform(10,22)
trips/day miles miles/hour hours Wh/mile
a Lambda (l) varies hourly based on a single peak model. The base value of l is 1 trip/hour at 7 AM. b Trip length was generated by the function of 10xBeta(2,6)þ1. c Speed source (Cherry and He, 2010). d Energy consumption source (Lemire-Elmore, 2004; Weinert et al., 2007b).
electricity to operate and battery management is essential to ensure high levels of availability. In order to make broad comparisons, we developed a Monte Carlo simulation model to evaluate the performance of e-bike sharing systems, focusing on individual stations. This versatile model can easily be extended to simulate different scenarios; for example, one-way trip-making patterns can be modeled by eliminating intra-trip activities. Our study focuses on three demand variables e trip generation, trip length, and trip duration. We also used different battery recharge rates to assess system ability if the kiosks were equipped with more expensive fast-charging systems (Göllei et al., 2012). The output of the simulation is the total number of e-bikes and batteries required to meet various levels of travel demand under different recharge rates. Each scenario was simulated 30 times using the same set of different random seeds. There are two modules for the simulation process shown in the flow chart (Fig. 3).Module (1) Trip Generation module (trip rate, trip length, and trip duration)Module (2) Event Process module (arrival, return, and recharging events) 3.2. Trip generation module
2.2. Operational concepts Under UTK’s e-bike sharing system, a subscriber checks out an e-bike by swiping a magnetic card to enter credentials at the kiosk. The system automatically releases a battery with at least 75% state of charge (SOC) and unlocks an e-bike for the subscriber, who then inserts the battery into the battery slot of the e-bike and commences the trip. Because of the limited number of stand-alone stations in this pilot study, users are allowed to check out e-bikes for up to 4 h without penalty (round trip model: Check-out / Activity / Check-in), compared to other bike sharing systems with ubiquitous station locations and short one-way trips (one-way trip model: Check-out / Check-in / Activity). Each e-bike comes with a combination lock, which the user may use at intermediate destinations. The system requires the user to re-enter credentials when returning the e-bike. The user will also return the battery at this time to the battery recharging bank in the kiosk. 3. E-bike sharing system simulation 3.1. Method Besides the UTK campus system, there are no known fully automated e-bike sharing systems in North America and perhaps only a few small-scale installations worldwide. Although some researchers expect this technology to grow rapidly, there is little published research or empirical data regarding the effects such systems could have on travel demand. E-bikes may introduce a number of operational complications into traditional bike sharing schemes. Unlike traditional bicycle sharing, e-bikes require
Table 1 lists input variables and statistical distributions for the trip generation module. In this module, trip demands were assumed to follow a Poisson distribution, commonly used to model the occurrence of rare events, or events with low probability of occurring during a fixed interval. The trip demands by the subscribers in the pilot study are less than 10 trips per day on average with little clustering. However, a fully deployed system is expected to have higher demand levels. Some fully deployed bike sharing systems see very high demand or significantly clustered demand. In those cases, statistical distributions other than Poisson would be more suitable. In the Poisson distribution, Lambda (l) is a parameter representing the expected number of occurrences in a unit of time. Hourly l values following a single-peak trip rate curve were selected in our study based on the field observed data on UTK campus. In the base case, the base hourly demand is 1.0 at 7:00. Then hourly demand rate increases progressively by 20% every hour until 13:00. From 13:00, the demand rate decreases by 20% every hour until 18:00. Considering the base-trip rate as described above, our e-bike sharing system yields about two trips per e-bike per day, on average. Round trip length for each trip is generated randomly based on a Beta distribution. The round trip length ranges from one mile to eleven miles with an average of 3.5 miles (Pucher and Buehler, 2006; Rietveld and Daniel, 2004). The shape of Beta distribution has a positive skew. The energy consumption rate is estimated by multiplying trip distance by battery discharge rate, which was assumed to follow a uniform distribution on the interval of 10e22 W-hours (Wh) per mile based on empirical measurements by the authors and previous research (Lemire-Elmore, 2004; Weinert et al., 2007b).
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8 Table 2 Performance analysis for e-bike sharing system (ten e-bikes and ten batteries). Trip ratea (trips/day)
Trip lengthb (mile)
Activity durationc (hour)
Charge Rated
Frequency of No bicyclee
Frequency of No batteryf
Service rateg (% served)
20
1.8
1
1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x
1 1 12 12 48 48 2 2 11 11 83 87 0 1 14 28 73 92
0 0 0 0 0 0 1 0 0 0 5 1 16 0 31 4 46 9
99.8% 99.8% 98.1% 98.1% 92.2% 92.2% 99.5% 99.7% 98.2% 98.2% 85.8% 85.8% 97.5% 99.8% 92.7% 94.8% 81.0% 83.9%
2 4 3.5
1 2 4
7
1 2 4
a On average, there is 20 trips per day for ten bicycles. Each simulation has on average 600 check out events (20 trips* 30 simulations). b Mean value for trip length. c Mean activity duration. d “1x” is slow-charging (4e6 h) and “2x” is quick-charging. e “No Bicycle” means the demand can’t be served since there are no more e-bikes in stock. This is the total number for 30 simulations. f “No Battery” means the demand can’t be served since the maximum energy of batteries in stock is less than 180 Wh. This is the total count for 30 simulations. g “Service Rate” means, under each scenario, the proportion of total trip demands for 30 simulations can be served by 10 bicycles and 10 batteries.
Trip duration includes two components: travel time, which is estimated by dividing trip length by riding speed, and activity duration. Riding speed (miles per hour) is assumed to be normally distributed (N(9, 4)) per Cherry and He (2010). Activity time (hour), during which the e-bike is not used for riding but is otherwise unavailable to other riders, is also assumed to be normally distributed (N(2, 2)). It should be noted that an e-bike and its onboard battery are unavailable for rental and recharging for the entire trip duration. 3.3. Processing events The Processing Events module includes three sub-modules e arrival (of a demand at a station), return (of the e-bike and battery), and recharging modules. The arrival event module is used to allocate the e-bike and battery for each trip demanded. For any given request, batteries with less than 75% (180 Wh) of the capacity charge will not be released to the user to guarantee a reasonable range of round-trip travel. On the other hand, for the purpose of efficiency, batteries do not have to be fully charged to be released, because it can take the same amount of time to charge a battery from 80% to 100% as it does from 0% to 80%. When the e-bike is returned (based on check-out time, travel time, and activity duration) the return event module will update the station status matrix, which records the number of e-bikes, number of batteries, and the battery recharge state. Based on the updated station matrix, the batteries are recharged under the recharging event module. In the recharge module, individual battery capacity is fixed at 240 Wh. Recharge rates vary depending on the amount of discharge. For a baseline, if a battery’s capacity is between 200 and 240 Wh, its recharge rate is 0.33 Wh per minute. If it is below 200 Wh, the recharge rate is a much higher 0.95 Wh per minute, until it reaches 200 Wh, based on the authors’ empirical observations of the recharge curve of batteries in the system.
5
4. Results Table 2 shows the simulation results of a single station system with ten e-bikes and ten batteries with a few combinations of average trip lengths, activity durations, and recharge rates. The average daily demand for the system is 20 trips. The results show that by applying quick-charging (reducing the recharge time to half of the baseline), the number of batteries out of service can be lowered dramatically. However, the service rate for the system does not increase substantially because of the limited number of available ebikes. If we lower the activity duration to 2 h on average, the system will satisfy at least 90% of trip demands. Therefore, a pre-defined duration for users to return the e-bikes is crucial to the system, balancing system performance, capital cost, and user convenience. 4.1. Sensitivity analysis To study the sensitivity of these results under various demand and supply scenarios, we evaluated demand levels deviating from the base case at 20% increments. The mean trip rate ranged from 60% to 200% of the base case, mean activity time ranged from 0.4 to 4.0 h. Mean trip length ranged from 2.1 to 5.6 miles, forming a demand variation matrix. The statistical distributions for the simulation are the same as those presented in Table 1. On the supply side, two battery-recharging rates were considered. A full factorial design yielded 960 unique combinations (8 demand levels 10 activity durations 6 trip lengths 2 charging rates) of scenarios. Each scenario was simulated 30 times (with random seeds), yielding 28,800 unique observations. Fig. 4 is a visual representation of simulation results for minimum number of e-bikes and batteries required for each run to meet maximum demand considering trip length and trip rate. The two sub-figures on the top level (Fig. 4a and b) are for the minimum number of e-bikes required, and the two sub-figures on the lower level (Fig. 4c and d) are for the minimum number of batteries required. These plots suggest a strong positive relationship between the minimum number of e-bikes/batteries required and the trip rate. Activity duration has impacts on the minimum number of e-bikes and the minimum number of batteries. A correlation between the minimum number of e-bikes/batteries and the trip length is less evident. If the demand and/or activity duration increased significantly, more e-bikes would be needed to maintain the maximum level of service. To corroborate the visual examination in Fig. 4, multiple linear regression analyses were performed to estimate the minimum number of batteries and e-bikes required to meet various demand levels under slow and quick recharging scenarios. The regression approach quantifies specific effects of activity duration and trip demand with the outcome variables, e-bike and battery availability. The results are shown in Table 3. The coefficients and signs are expected (by design). Activity duration has a strong effect on the number of e-bikes and batteries needed to serve the demand. Reducing activity duration by 1 h has the equivalent effect on e-bike and battery requirements as increasing trip rate by a factor of six, all else equal. Battery requirements are about twice as sensitive to trip length as e-bike fleet requirements, since e-bike battery availability (inversely related to required recharge time) is linked to trip length. An application of this model could be to design the size of a station based on demand variables that one might observe or derive. For illustration, consider a station that has a trip rate of 40 trips per day, average round trip length of six miles, activity duration of 2 h, and a standard battery recharge rate. These models predict that this station would require 16 e-bikes and 18 batteries to meet the demand. If a quick charging mechanism is used instead,
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
6
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
Fig. 4. Minimum number of e-bikes and batteries required to meet demand.
the number of batteries would reduce from 18 to 17, just one more than the minimum e-bikes available. An important note is that the battery requirements are constrained by e-bike requirements. That is, one needs at least as many batteries as e-bikes. The implication is Table 3 Regression estimate of minimum e-bike and battery requirements. Minimum E-bikes required (R2 ¼ 0.83) Parameter estimate
Minimum batteries required (quick charge) (R2 ¼ 0.84)
Minimum batteries required (slow charge) (R2 ¼ 0.84) pParameter value estimate
Intercept 5.48 (0.085) 0.00 Trip rate 0.37 (0.002) 0.00 (trips/day) Trip length 0.37 (0.016) 0.00 (mile) 2.43 (0.016) 0.00 Activity duration (hour) Note: Standard error in parenthesis.
pParameter value estimate
pvalue
6.87 (0.086) 0.00 0.39 (0.002) 0.00
6.21 (0.085) 0.00 0.38 (0.002) 0.00
0.76 (0.016) 0.00
0.58 (0.016) 0.00
2.36 (0.017) 0.00
2.39 (0.016) 0.00
that one would require battery switching to meet maximum demand (or the station will have available e-bikes, but no batteries). Furthermore, based on the simulation results, the difference in minimum batteries required under slow charging or fast charging scenarios varies from 0 to 4. For 25% of the simulations, the system can only reduce battery requirements by one by switching to a fast charging operation. One should carefully consider the costs and benefits of using a more expensive fast charging mechanism or to increase the battery inventory. These Monte Carlo simulations can also be employed for assessing one-way trip scenarios by eliminating activity duration. Since activity duration dominates our model in a one-station round-trip based system, removing vehicle out-of-service time by allowing one-way use would substantially reduce vehicle fleet requirements, but introduces other fleet redistribution challenges in a multi-station e-bike sharing system. 5. Conclusion The UTK campus is ideal for an e-bike sharing system because the campus is hilly and geographically spread out, resulting in a
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
large potential pool of customers that might consider using an ebike, but not necessarily a bicycle. Based on the simulation results, we found that the minimum number of e-bikes and batteries required by the system are significantly sensitive to trip rate, trip length, and activity duration. E-bike and battery inventory is very sensitive to the trip rate. Activity duration also has a strong impact on the system service rate. Thus, measures should be taken to incentivize efficient use of ebikes, perhaps in the form of pricing strategies to inhibit high duration trips (similar to pricing schemes of existing bike sharing systems). The effect of trip length on battery availability is minor compared to activity duration. The need to recharge batteries requires access to a reliable energy source. The most reliable source is a connection to an electric utility. However, this complicates selection of station sites and could increase installation cost. The need to store solar energy in batteries or other devices is a common disadvantage of solar-based systems, primarily due to the expense of the storage components. Since battery recharging is the primary power requirement in this application and the batteries are an integral component of the system, an e-bike sharing system may be an appropriate application for solar power (Redpath et al., 2011). Nevertheless, sufficient reserve energy capacity may be crucial to uninterrupted service in periods of inclement weather or limited daylight. The fact that biking activity and energy production are somewhat correlated during inclement weather may offset this limitation. Hybrid grid and solar power would provide both uninterrupted service and the benefits of a renewable energy source. Because e-bike batteries are swappable, solar power can charge the e-bike batteries directly; potentially limiting the energy loss when recharging e-bike batteries with large storage batteries typically used in solar power systems. Also, optimized size of solar panel and capacity of storage device should be selected. This simulation can be extended to estimate the capacity and design of solar-powered e-bike sharing stations and intelligent load management could require more sophisticated stock, charging, and distribution of e-bike batteries in a solar-power only system. As bike sharing grows as an important mode in urban transportation systems, shared e-bikes could find a significant niche in our transportation system. E-bikes expand potential markets to those who are not inclined to traditional bicycling. Shifts from other motorized modes reduce energy use and emissions while moving more individuals toward active transport modes. More studies with real-world data are needed, for instance, to compare costs of using e-bikes to costs of other modes, determine the costs and benefits of providing more expensive quick-chargers, evaluate the impacts on the environment and energy savings, and assess impacts on physical activity and public health. Further empirical studies are planned, using empirical user data. Acknowledgments This project was funded by the Southeastern Transportation Center, Tennessee Department of Transportation, UT’s Make Orange Green sustainability initiative, and the National Science Foundation under grant CBET-1055282. The authors would like to thank Stacy Worley, David Smith, John Wilkerson, Casey Langford, Taekwan Yoon, Bryanna Shelton, and Larry Roberts for their support in developing the system. References Angeloudis, P., Hu, J., Bell, M.G.H., 2012. A strategic repositioning algorithm for bicycle-sharing schemes. In: Transportation Research Board 91st Annual Meeting, Washington DC.
7
Bachand-Marleau, J., Lee, B., El-Geneidy, A., 2012. Towards a better understanding of the factors influencing the likelihood of using shared bicycle systems and frequency of use. Transport. Res. Record: J. Transport. Res. Board 2314, 66e71. Buck, D., Buehler, R., 2012. Bike lanes and other determinants of capital bikeshare trips. In: Transportation Research Board 91st Annual Meeting, Washington DC. Cherry, C., He, M., 2010. Alternative methods of measuring operating speed of electric and traditional bikes in China-implications for travel demand models. J. East. Asia Soc. Transport. Stud. 8, 1424e1436. Cherry, C.R., Weinert, J.X., Xinmiao, Y., 2009. Comparative environmental impacts of electric bikes in China. Transport. Res. Part D: Trans. Environ. 14, 281e290. DeMaio, P., 2009. Bike-sharing: history, impacts, models of provision, and future. J. Pub. Transport. 12, 41e56. Dill, J., Rose, G., 2012. E-Bikes and transportation policy: insights from early adopters. In: Transportation Research Board 91st Annual Meeting, Washington DC. García-Palomares, J.C., Gutiérrez, J., Latorre, M., 2012. Optimizing the location of stations in bike-sharing programs: a GIS approach. Appl. Geogr. 35, 235e246. Gleason, R., Miskimins, L., 2012. Exploring Bicycle Options for Federal Lands: Bike Sharing, Rentals and Employee Fleets. Western Transportation Institute. Gojanovic, B., Welker, J., Iglesias, K., Daucourt, C., Gremion, G., 2011. Electric bicycles as a new active transportation modality to promote health. Med. Sci. Sports Exerc. 43, 2204e2210. Göllei, A., Görbe, P., Magyar, A., 2012. Modeling and optimization of electrical vehicle batteries in complex clean energy systems. J. Clean. Prod. 34, 138e145. Hampshire, R.C., Marla, L., 2012. An analysis of bike sharing usage: explaining trip generation and attraction from observed demand. In: Transportation Research Board 91st Annual Meeting, Washington DC. Ji, S., Cherry, C.R., Bechle, M.J., Wu, Y., Marshall, J.D., 2012. Electric vehicles in China: emissions and health impacts. Environ. Sci. Technol. 46, 2018e2024. Kaltenbrunner, A., Meza, R., Grivolla, J., Codina, J., Banchs, R., 2010. Urban cycles and mobility patterns: exploring and predicting trends in a bicycle-based public transport system. Per. Mob. Comp. 6, 455e466. Kim, D., Shin, H., Im, H., Park, J., 2012. Factors influencing travel behaviors in bikesharing. In: Transportation Research Board 91st Annual Meeting, Washington DC. Krykewycz, G., Puchalsky, C., Rocks, J., Bonnette, B., Jaskiewicz, F., 2010. Defining a primary market and estimating demand for major bicycle-sharing program in Philadelphia, Pennsylvania. Transport. Res. Record: J. Transport. Res. Board 2143, 117e124. Langford, B., Cherry, C., Yoon, T., Worley, S., Smith, D., 2013. North America’s first ebike share: a year of experience. Transport. Res. Record: J. Transport. Res. Board (In Press). Lemire-Elmore, J., 2004. The Energy Cost of Electric and Human-powered Bicycles. Lin, J.-R., Yang, T.-H., 2011. Strategic design of public bicycle sharing systems with service level constraints. Transport. Res. Part E: Log. Transport. Rev. 47, 284e294. Louis, J., Brisswalter, J., Morio, C., Barla, C., Temprado, J.-J., 2012. The electrically assisted bicycle: an alternative way to promote physical activity. Am. J. Phys. Med. Rehab. 91, 931e940. Maurer, L.K., 2012. Feasibility study for a bicycle sharing program in Sacramento, California. In: Transportation Research Board 91st Annual Meeting, Washington DC. Nair, R., Miller-Hooks, E., Hampshire, R.C., Busi c, A., 2012. Large-scale vehicle sharing systems: analysis of Vélib. Int. J. Sust. Transport. 7, 85e106. Parkes, S., Marsden, G., 2012. City bike hire schemes e emerging trends in Europe. In: Transportation Research Board 91st Annual Meeting, Washington DC. Pucher, J., Buehler, R., 2006. Why Canadians cycle more than Americans: a comparative analysis of bicycling trends and policies. Trans. Policy 13, 265e279. Redpath, D.A.G., McIlveen-Wright, D., Kattakayam, T., Hewitt, N.J., Karlowski, J., Bardi, U., 2011. Battery powered electric vehicles charged via solar photovoltaic arrays developed for light agricultural duties in remote hilly areas in the Southern Mediterranean region. J. Clean. Prod. 19, 2034e2048. Rietveld, P., Daniel, V., 2004. Determinants of bicycle use: do municipal policies matter? Transport. Res. Part A: Policy Pract. 38, 531e550. Romero, J.P., Moura, J.L., Ibeas, A., Benavente, J., 2012. Car-bicycle combined model for planning bicycle sharing systems. In: Transportation Research Board 91st Annual Meeting, Washington DC. Rose, G., 2012. E-bikes and urban transportation: emerging issues and unresolved questions. Transportation 39, 81e96. Schroeder, B., Hagen, J., Leve, Z., Penalosa, A., 2009. Bike-sharing goes viral. Sust. Trans., 24e29. Shaheen, S., Cohen, A., Martin, E., 2013. Public bikesharing in North America: early operator understanding and emerging trends. Transport. Res. Record: J. Transport. Res. Board (forthcoming). Shaheen, S., Guzman, S., Zhang, H., 2010. Bikesharing in Europe, the Americas, and Asia. Transport. Res. Record: J. Transport. Res. Board 2143, 159e167. Shang, W., Jiangman, Z., Liang, L., Zheng-yu, D., 2010. Bike-sharing e a new public transportation mode: state of the practice & prospects. In: Emergency Management and Management Sciences (ICEMMS), 2010 IEEE International Conference on, pp. 222e225. Silvester, S., Beella, S.K., van Timmeren, A., Bauer, P., Quist, J., van Dijk, S., 2013. Exploring design scenarios for large-scale implementation of electric vehicles; the Amsterdam Airport Schiphol case. J. Clean. Prod. 48, 211e219. Sperlich, B., Zinner, C., Hébert-Losier, K., Born, D.-P., Holmberg, H.-C., 2012. Biomechanical, cardiorespiratory, metabolic and perceived responses to electrically assisted cycling. Eur. J. Appl. Physiol. 112, 4015e4025.
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
8
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
Theurel, J., Theurel, A., Lepers, R., 2012. Physiological and cognitive responses when riding an electrically assisted bicycle versus a classical bicycle. Ergonomics 55, 773e781. Tonn, B.E., Schexnayder, S.M., Peretz, J.H., Das, S., Waidley, G., 2003. An assessment of waste issues associated with the production of new, lightweight, fuelefficient vehicles. J. Clean. Prod. 11, 753e765.
Weinert, J., Ma, C., Cherry, C., 2007a. The transition to electric bikes in China: history and key reasons for rapid growth. Transportation 34, 301e318. Weinert, J.X., Burke, A.F., Wei, X., 2007b. Lead-acid and lithium-ion batteries for the Chinese electric bike market and implications on future technology advancement. J. Power Sour. 172, 938e945.
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
Contents lists available at ScienceDirect
Journal of Cleaner Production journal homepage: www.elsevier.com/locate/jclepro
Electric bike sharing: simulation of user demand and system availability Shuguang Ji a, Christopher R. Cherry b, *, Lee D. Han b, David A. Jordan b a b
University of Tennessee, Department of Industrial and Systems Engineering, Knoxville, TN, USA University of Tennessee, Department of Civil and Environmental Engineering, Knoxville, TN, USA
a r t i c l e i n f o
a b s t r a c t
Article history: Received 14 May 2013 Received in revised form 18 September 2013 Accepted 20 September 2013 Available online xxx
This paper describes the operational concepts and system requirements of a fully automated electric bike (e-bike) sharing system demonstrated through a pilot project at the University of Tennessee, Knoxville (UTK) campus (deployed in September 2011). This project is part of a movement to develop a sustainable transportation system, and is one of the green initiatives on UTK campus. E-bikes are more energy efficient and produce fewer greenhouse gas (GHG) emissions per person compared to other transport modes such as car, bus, and motorcycle. Without empirical demand information for an e-bike sharing system, we simulated the operations of such a system to gain insights during the design process before field deployment. The simulation exercise focused on three critical demand parameters e distributions of trip rates, trip lengths, and trip durations e and coupled them with supply parameters e number of ebikes, number of swappable batteries, and battery recharging profiles. The primary purpose of these simulations is to evaluate the efficiency of an off-board battery recharging system, where the depleted battery is removed from an e-bike upon its return and inserted into one of the charging slots at the kiosk. We tested various scenarios with different number of batteries always maintaining an initial condition with the battery to e-bike ratio greater or equal to 1.0 to ensure battery availability. We applied empirical battery recharging rates and system operations rules to determine the number of e-bikes and batteries available under different potential demand situations, with a focus on optimizing the number of batteries to meet user demands. By adjusting input parameters, numerous scenarios were simulated for sensitivity analysis. Based on the results of the simulation, this paper presents a cost constrained e-bike sharing system design that can maintain a high level of system reliability (e-bike and battery availability) through optimal battery charging and distribution management. We found that high demand scenarios require multiple swappable batteries per e-bike to reasonably meet the maximum demand. Trip duration has the most influence on e-bike and battery availability, followed by trip rate, and then trip length. Ó 2013 Elsevier Ltd. All rights reserved.
Keywords: Electric bike sharing System operations Monte Carlo simulation E-bike
1. Introduction Bike sharing is a new provision model of public/private transportation that has emerged in the past several years. As of January 2012, there are 15 automated bike sharing programs in the U.S. involving 5238 bicycles and 172,070 users (Shaheen et al., 2013). Bike sharing is among the most environmentally sound models of transport that provides public transportation services for short trips that are often inadequately served by other modes of fixedroute public transit (Hampshire and Marla, 2012; Shang et al., 2010). Bike sharing systems have been overcoming many operational challenges in the past decade to provide fully automated,
* Corresponding author. Tel.: þ1 865 974 7710; fax: þ1 865 974 2669. E-mail address: [email protected] (C.R. Cherry).
secure, and cost-effective systems and have been deployed in over 100 cities worldwide (Kim et al., 2012). Traditional bicycles in bike sharing systems range in price, with mainstream systems offering bicycles for about US$1000 (Shaheen et al., 2013). Electric bikes (ebikes) range in price as well but typically carry a price premium of approximately 2 for a similarly equipped bicycle. Formal bike sharing systems have existed for nearly half a century, with various levels of success. Few studies have systematically estimated demand or defined operational parameters (Kaltenbrunner et al., 2010). There have been three generations of evolution, driven mostly by advances in technology. The first generation began in Amsterdam in 1965, where stationless bicycles could be borrowed and left anywhere in the city, to be borrowed again by the next individual. It was unsuccessful due to vandalism and theft. The second generation, born in Denmark in 1991, allowed bicycles to be picked up and returned to several central locations
0959-6526/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jclepro.2013.09.024
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
2
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
Fig. 1. Typical pedal e-bike.
with a coin deposit. Theft was also a problem largely due to the anonymity of the user. Third generation bike sharing was born in Portsmouth University in England and involved several technological improvements such as bike racks that locked automatically, on-board electronics, swipe cards, and telecommunication capabilities. In 2005 and 2007 respectively, Lyon and Paris, France launched highly successful third generation bike sharing programs that grew to over 15,000 and 20,000 bicycles respectively. Today, one bike sharing program per month is being created somewhere in the world (DeMaio, 2009; Shaheen et al., 2010). Beginning in 2008, cities outside of Europe began to launch third generation programs. Rio de Janeiro launched a pilot public bike sharing program in 2009. Several others followed in South America and Asia. Some of the largest bike sharing networks are in cities such as Hangzhou (40,000 bicycles and 1700 stations) and Wuhan, China (13,000 bicycles and 516 stations). In North America, Montreal, Denver, Minneapolis, Boston, and Washington D.C. have launched successful third generation bike sharing programs in recent years. Dozens of other North American cities are in planning stages for installing bike sharing systems (Krykewycz et al., 2010; Schroeder et al., 2009). Simultaneously, e-bikes have gained popularity in many regions of the world. Some researchers suggest that shared e-bikes could provide an even higher level of service compared to existing bike
sharing systems, while maintaining low environmental impacts. Based on recent research findings, benefits such as the ability to travel longer distances and over hills with less fatigue and sweat, compared to traditional bicycles, may help overcome some of the barriers to bicycling. In addition, surveys of e-bike riders reveal that most e-bike owners favor e-bikes over traditional bicycles or conventional motorized vehicles (Dill and Rose, 2012). E-bikes are often regulated like bicycles in most jurisdictions. Ebikes, particularly pedal assist e-bikes, look and operate much like traditional bicycles (Fig. 1). Pedal assist e-bikes offer some of the health benefits of bicycling, requiring relatively modest amounts of physical activity (Gojanovic et al., 2011; Louis et al., 2012; Sperlich et al., 2012; Theurel et al., 2012). However, e-bikes are substantially more expensive than similar-quality non-electric bicycles. As such, the e-bike market has not grown as rapidly in the U.S. compared to other countries (Dill and Rose, 2012; Rose, 2012; Weinert et al., 2007a). By spreading the higher cost of e-bikes over a community of many users and over time, e-bike sharing has the potential of overcoming the price barrier as well as the perceived expense of the technology. The provision of e-bikes in a shared environment also introduces electric vehicle technology to potential users without the pressure or commitment of capital purchase. E-bike sharing promotes electric mobility to users of other motorized modes of transport and introduces renewable energy, recharging infrastructure, and safety to the public (Silvester et al., 2013). By affecting mode shifts of users from other motorized modes and because of their light weight, e-bikes are considered a technology towards cleaner transportation (Tonn et al., 2003). E- bikes outperform other motorized transport modes, including bus transit, on energy efficiency and greenhouse gas emission (GHG) rates per passenger kilometer, considering the complete life cycle environmental impacts (Cherry et al., 2009). Moreover, e-bikes can move emissions from tailpipes to power plants that are often away from urban areas, further reducing human exposures to emissions (Ji et al., 2012). Traditional bike sharing systems have been extensively researched in the past few years. Many studies aim to provide policy recommendations to encourage more bicycle trips (Bachand-Marleau et al., 2012; Gleason and Miskimins, 2012; Parkes and Marsden, 2012). Some studies try to employ geographic information system (GIS) to estimate the demand for
Fig. 2. System components of e-bike sharing system.
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
3
Fig. 3. Flow chart of simulation process.
the bike sharing programs (García-Palomares et al., 2012; Hampshire and Marla, 2012; Krykewycz et al., 2010) or develop new models for the configuration of bike sharing systems (Angeloudis et al., 2012; Kim et al., 2012; Lin and Yang, 2011; Maurer, 2012; Nair et al., 2012). Some studies investigate the influences of bicycle infrastructure (e.g., bicycle lanes) on the bike sharing system demand (Buck and Buehler, 2012; Romero et al., 2012). In this paper, we attempt to extend this bike sharing research domain by incorporating e-bikes into the system, focusing
on system configuration, and managing supply to meet demand through a simulation approach. 2. E-bike sharing system description 2.1. System components In September 2011, the University of Tennessee-Knoxville (UTK) installed North America’s first e-bike sharing station as a technical
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
4
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
and operational research pilot test (Langford et al., 2013). This pilot test platform has two stand-alone stations with ten bicycles each and off-board battery charging capability for battery swapping. Each station has the capacity to charge and distribute up to 15 batteries at a time. The e-bike sharing system consists of pedal assist e-bikes, powered by lithium ion (Li-ion) batteries (Fig. 1), and a vending and charging station (Fig. 2). Each pedal assist e-bike has a sensor measuring the rider’s pedaling effort continuously. This information is used by an on-board control device to supplement the rider’s effort with electromechanical power when necessary to help the rider overcome steep hills and long grades without eliminating the need for the rider to pedal. Increased rider effort increases the electric-assist range and reduces fatigue barriers while still providing some health benefits to the rider and maintaining the legal and perceived identity of e-bike as a bicycle. These benefits make the system and vehicles more attractive to casual riders, who may not otherwise consider traditional bicycles as a viable transport mode. Li-ion batteries are the most common in North American and European e-bikes, primarily because of a relatively low weight and high energy density. To extend battery life, maximize stored energy, and minimize the risk of overcharging and overheating, it is advisable for an e-bike sharing system to adopt a slower battery charging system, which also reduces the availability of batteries. Physical security, vending, and charging are the three main functions of the station, which is equipped with an electromechanical locking system that simplifies the check-in/check-out process and verifies when an e-bike is returned and properly secured. The e-bike sharing station provides access to registered users only. Users are required to remove and place the used battery for recharging as a part of the check-in procedure. While this extra step adds user burden, it maximizes recharging activities automatically, independent of the availability of the e-bike.
Table 1 Input variables and statistical distributions for Monte Carlo simulation. Variable
Base-case value
Statistical distribution model
Units
Trip rate Trip length Riding speedc Activity duration Energy consumption rated
20 3.5 9 2 16
Poisson(l)a Beta(2,6)b Normal(9,4) Normal(2,2) Uniform(10,22)
trips/day miles miles/hour hours Wh/mile
a Lambda (l) varies hourly based on a single peak model. The base value of l is 1 trip/hour at 7 AM. b Trip length was generated by the function of 10xBeta(2,6)þ1. c Speed source (Cherry and He, 2010). d Energy consumption source (Lemire-Elmore, 2004; Weinert et al., 2007b).
electricity to operate and battery management is essential to ensure high levels of availability. In order to make broad comparisons, we developed a Monte Carlo simulation model to evaluate the performance of e-bike sharing systems, focusing on individual stations. This versatile model can easily be extended to simulate different scenarios; for example, one-way trip-making patterns can be modeled by eliminating intra-trip activities. Our study focuses on three demand variables e trip generation, trip length, and trip duration. We also used different battery recharge rates to assess system ability if the kiosks were equipped with more expensive fast-charging systems (Göllei et al., 2012). The output of the simulation is the total number of e-bikes and batteries required to meet various levels of travel demand under different recharge rates. Each scenario was simulated 30 times using the same set of different random seeds. There are two modules for the simulation process shown in the flow chart (Fig. 3).Module (1) Trip Generation module (trip rate, trip length, and trip duration)Module (2) Event Process module (arrival, return, and recharging events) 3.2. Trip generation module
2.2. Operational concepts Under UTK’s e-bike sharing system, a subscriber checks out an e-bike by swiping a magnetic card to enter credentials at the kiosk. The system automatically releases a battery with at least 75% state of charge (SOC) and unlocks an e-bike for the subscriber, who then inserts the battery into the battery slot of the e-bike and commences the trip. Because of the limited number of stand-alone stations in this pilot study, users are allowed to check out e-bikes for up to 4 h without penalty (round trip model: Check-out / Activity / Check-in), compared to other bike sharing systems with ubiquitous station locations and short one-way trips (one-way trip model: Check-out / Check-in / Activity). Each e-bike comes with a combination lock, which the user may use at intermediate destinations. The system requires the user to re-enter credentials when returning the e-bike. The user will also return the battery at this time to the battery recharging bank in the kiosk. 3. E-bike sharing system simulation 3.1. Method Besides the UTK campus system, there are no known fully automated e-bike sharing systems in North America and perhaps only a few small-scale installations worldwide. Although some researchers expect this technology to grow rapidly, there is little published research or empirical data regarding the effects such systems could have on travel demand. E-bikes may introduce a number of operational complications into traditional bike sharing schemes. Unlike traditional bicycle sharing, e-bikes require
Table 1 lists input variables and statistical distributions for the trip generation module. In this module, trip demands were assumed to follow a Poisson distribution, commonly used to model the occurrence of rare events, or events with low probability of occurring during a fixed interval. The trip demands by the subscribers in the pilot study are less than 10 trips per day on average with little clustering. However, a fully deployed system is expected to have higher demand levels. Some fully deployed bike sharing systems see very high demand or significantly clustered demand. In those cases, statistical distributions other than Poisson would be more suitable. In the Poisson distribution, Lambda (l) is a parameter representing the expected number of occurrences in a unit of time. Hourly l values following a single-peak trip rate curve were selected in our study based on the field observed data on UTK campus. In the base case, the base hourly demand is 1.0 at 7:00. Then hourly demand rate increases progressively by 20% every hour until 13:00. From 13:00, the demand rate decreases by 20% every hour until 18:00. Considering the base-trip rate as described above, our e-bike sharing system yields about two trips per e-bike per day, on average. Round trip length for each trip is generated randomly based on a Beta distribution. The round trip length ranges from one mile to eleven miles with an average of 3.5 miles (Pucher and Buehler, 2006; Rietveld and Daniel, 2004). The shape of Beta distribution has a positive skew. The energy consumption rate is estimated by multiplying trip distance by battery discharge rate, which was assumed to follow a uniform distribution on the interval of 10e22 W-hours (Wh) per mile based on empirical measurements by the authors and previous research (Lemire-Elmore, 2004; Weinert et al., 2007b).
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8 Table 2 Performance analysis for e-bike sharing system (ten e-bikes and ten batteries). Trip ratea (trips/day)
Trip lengthb (mile)
Activity durationc (hour)
Charge Rated
Frequency of No bicyclee
Frequency of No batteryf
Service rateg (% served)
20
1.8
1
1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x 1x 2x
1 1 12 12 48 48 2 2 11 11 83 87 0 1 14 28 73 92
0 0 0 0 0 0 1 0 0 0 5 1 16 0 31 4 46 9
99.8% 99.8% 98.1% 98.1% 92.2% 92.2% 99.5% 99.7% 98.2% 98.2% 85.8% 85.8% 97.5% 99.8% 92.7% 94.8% 81.0% 83.9%
2 4 3.5
1 2 4
7
1 2 4
a On average, there is 20 trips per day for ten bicycles. Each simulation has on average 600 check out events (20 trips* 30 simulations). b Mean value for trip length. c Mean activity duration. d “1x” is slow-charging (4e6 h) and “2x” is quick-charging. e “No Bicycle” means the demand can’t be served since there are no more e-bikes in stock. This is the total number for 30 simulations. f “No Battery” means the demand can’t be served since the maximum energy of batteries in stock is less than 180 Wh. This is the total count for 30 simulations. g “Service Rate” means, under each scenario, the proportion of total trip demands for 30 simulations can be served by 10 bicycles and 10 batteries.
Trip duration includes two components: travel time, which is estimated by dividing trip length by riding speed, and activity duration. Riding speed (miles per hour) is assumed to be normally distributed (N(9, 4)) per Cherry and He (2010). Activity time (hour), during which the e-bike is not used for riding but is otherwise unavailable to other riders, is also assumed to be normally distributed (N(2, 2)). It should be noted that an e-bike and its onboard battery are unavailable for rental and recharging for the entire trip duration. 3.3. Processing events The Processing Events module includes three sub-modules e arrival (of a demand at a station), return (of the e-bike and battery), and recharging modules. The arrival event module is used to allocate the e-bike and battery for each trip demanded. For any given request, batteries with less than 75% (180 Wh) of the capacity charge will not be released to the user to guarantee a reasonable range of round-trip travel. On the other hand, for the purpose of efficiency, batteries do not have to be fully charged to be released, because it can take the same amount of time to charge a battery from 80% to 100% as it does from 0% to 80%. When the e-bike is returned (based on check-out time, travel time, and activity duration) the return event module will update the station status matrix, which records the number of e-bikes, number of batteries, and the battery recharge state. Based on the updated station matrix, the batteries are recharged under the recharging event module. In the recharge module, individual battery capacity is fixed at 240 Wh. Recharge rates vary depending on the amount of discharge. For a baseline, if a battery’s capacity is between 200 and 240 Wh, its recharge rate is 0.33 Wh per minute. If it is below 200 Wh, the recharge rate is a much higher 0.95 Wh per minute, until it reaches 200 Wh, based on the authors’ empirical observations of the recharge curve of batteries in the system.
5
4. Results Table 2 shows the simulation results of a single station system with ten e-bikes and ten batteries with a few combinations of average trip lengths, activity durations, and recharge rates. The average daily demand for the system is 20 trips. The results show that by applying quick-charging (reducing the recharge time to half of the baseline), the number of batteries out of service can be lowered dramatically. However, the service rate for the system does not increase substantially because of the limited number of available ebikes. If we lower the activity duration to 2 h on average, the system will satisfy at least 90% of trip demands. Therefore, a pre-defined duration for users to return the e-bikes is crucial to the system, balancing system performance, capital cost, and user convenience. 4.1. Sensitivity analysis To study the sensitivity of these results under various demand and supply scenarios, we evaluated demand levels deviating from the base case at 20% increments. The mean trip rate ranged from 60% to 200% of the base case, mean activity time ranged from 0.4 to 4.0 h. Mean trip length ranged from 2.1 to 5.6 miles, forming a demand variation matrix. The statistical distributions for the simulation are the same as those presented in Table 1. On the supply side, two battery-recharging rates were considered. A full factorial design yielded 960 unique combinations (8 demand levels 10 activity durations 6 trip lengths 2 charging rates) of scenarios. Each scenario was simulated 30 times (with random seeds), yielding 28,800 unique observations. Fig. 4 is a visual representation of simulation results for minimum number of e-bikes and batteries required for each run to meet maximum demand considering trip length and trip rate. The two sub-figures on the top level (Fig. 4a and b) are for the minimum number of e-bikes required, and the two sub-figures on the lower level (Fig. 4c and d) are for the minimum number of batteries required. These plots suggest a strong positive relationship between the minimum number of e-bikes/batteries required and the trip rate. Activity duration has impacts on the minimum number of e-bikes and the minimum number of batteries. A correlation between the minimum number of e-bikes/batteries and the trip length is less evident. If the demand and/or activity duration increased significantly, more e-bikes would be needed to maintain the maximum level of service. To corroborate the visual examination in Fig. 4, multiple linear regression analyses were performed to estimate the minimum number of batteries and e-bikes required to meet various demand levels under slow and quick recharging scenarios. The regression approach quantifies specific effects of activity duration and trip demand with the outcome variables, e-bike and battery availability. The results are shown in Table 3. The coefficients and signs are expected (by design). Activity duration has a strong effect on the number of e-bikes and batteries needed to serve the demand. Reducing activity duration by 1 h has the equivalent effect on e-bike and battery requirements as increasing trip rate by a factor of six, all else equal. Battery requirements are about twice as sensitive to trip length as e-bike fleet requirements, since e-bike battery availability (inversely related to required recharge time) is linked to trip length. An application of this model could be to design the size of a station based on demand variables that one might observe or derive. For illustration, consider a station that has a trip rate of 40 trips per day, average round trip length of six miles, activity duration of 2 h, and a standard battery recharge rate. These models predict that this station would require 16 e-bikes and 18 batteries to meet the demand. If a quick charging mechanism is used instead,
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
6
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
Fig. 4. Minimum number of e-bikes and batteries required to meet demand.
the number of batteries would reduce from 18 to 17, just one more than the minimum e-bikes available. An important note is that the battery requirements are constrained by e-bike requirements. That is, one needs at least as many batteries as e-bikes. The implication is Table 3 Regression estimate of minimum e-bike and battery requirements. Minimum E-bikes required (R2 ¼ 0.83) Parameter estimate
Minimum batteries required (quick charge) (R2 ¼ 0.84)
Minimum batteries required (slow charge) (R2 ¼ 0.84) pParameter value estimate
Intercept 5.48 (0.085) 0.00 Trip rate 0.37 (0.002) 0.00 (trips/day) Trip length 0.37 (0.016) 0.00 (mile) 2.43 (0.016) 0.00 Activity duration (hour) Note: Standard error in parenthesis.
pParameter value estimate
pvalue
6.87 (0.086) 0.00 0.39 (0.002) 0.00
6.21 (0.085) 0.00 0.38 (0.002) 0.00
0.76 (0.016) 0.00
0.58 (0.016) 0.00
2.36 (0.017) 0.00
2.39 (0.016) 0.00
that one would require battery switching to meet maximum demand (or the station will have available e-bikes, but no batteries). Furthermore, based on the simulation results, the difference in minimum batteries required under slow charging or fast charging scenarios varies from 0 to 4. For 25% of the simulations, the system can only reduce battery requirements by one by switching to a fast charging operation. One should carefully consider the costs and benefits of using a more expensive fast charging mechanism or to increase the battery inventory. These Monte Carlo simulations can also be employed for assessing one-way trip scenarios by eliminating activity duration. Since activity duration dominates our model in a one-station round-trip based system, removing vehicle out-of-service time by allowing one-way use would substantially reduce vehicle fleet requirements, but introduces other fleet redistribution challenges in a multi-station e-bike sharing system. 5. Conclusion The UTK campus is ideal for an e-bike sharing system because the campus is hilly and geographically spread out, resulting in a
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
large potential pool of customers that might consider using an ebike, but not necessarily a bicycle. Based on the simulation results, we found that the minimum number of e-bikes and batteries required by the system are significantly sensitive to trip rate, trip length, and activity duration. E-bike and battery inventory is very sensitive to the trip rate. Activity duration also has a strong impact on the system service rate. Thus, measures should be taken to incentivize efficient use of ebikes, perhaps in the form of pricing strategies to inhibit high duration trips (similar to pricing schemes of existing bike sharing systems). The effect of trip length on battery availability is minor compared to activity duration. The need to recharge batteries requires access to a reliable energy source. The most reliable source is a connection to an electric utility. However, this complicates selection of station sites and could increase installation cost. The need to store solar energy in batteries or other devices is a common disadvantage of solar-based systems, primarily due to the expense of the storage components. Since battery recharging is the primary power requirement in this application and the batteries are an integral component of the system, an e-bike sharing system may be an appropriate application for solar power (Redpath et al., 2011). Nevertheless, sufficient reserve energy capacity may be crucial to uninterrupted service in periods of inclement weather or limited daylight. The fact that biking activity and energy production are somewhat correlated during inclement weather may offset this limitation. Hybrid grid and solar power would provide both uninterrupted service and the benefits of a renewable energy source. Because e-bike batteries are swappable, solar power can charge the e-bike batteries directly; potentially limiting the energy loss when recharging e-bike batteries with large storage batteries typically used in solar power systems. Also, optimized size of solar panel and capacity of storage device should be selected. This simulation can be extended to estimate the capacity and design of solar-powered e-bike sharing stations and intelligent load management could require more sophisticated stock, charging, and distribution of e-bike batteries in a solar-power only system. As bike sharing grows as an important mode in urban transportation systems, shared e-bikes could find a significant niche in our transportation system. E-bikes expand potential markets to those who are not inclined to traditional bicycling. Shifts from other motorized modes reduce energy use and emissions while moving more individuals toward active transport modes. More studies with real-world data are needed, for instance, to compare costs of using e-bikes to costs of other modes, determine the costs and benefits of providing more expensive quick-chargers, evaluate the impacts on the environment and energy savings, and assess impacts on physical activity and public health. Further empirical studies are planned, using empirical user data. Acknowledgments This project was funded by the Southeastern Transportation Center, Tennessee Department of Transportation, UT’s Make Orange Green sustainability initiative, and the National Science Foundation under grant CBET-1055282. The authors would like to thank Stacy Worley, David Smith, John Wilkerson, Casey Langford, Taekwan Yoon, Bryanna Shelton, and Larry Roberts for their support in developing the system. References Angeloudis, P., Hu, J., Bell, M.G.H., 2012. A strategic repositioning algorithm for bicycle-sharing schemes. In: Transportation Research Board 91st Annual Meeting, Washington DC.
7
Bachand-Marleau, J., Lee, B., El-Geneidy, A., 2012. Towards a better understanding of the factors influencing the likelihood of using shared bicycle systems and frequency of use. Transport. Res. Record: J. Transport. Res. Board 2314, 66e71. Buck, D., Buehler, R., 2012. Bike lanes and other determinants of capital bikeshare trips. In: Transportation Research Board 91st Annual Meeting, Washington DC. Cherry, C., He, M., 2010. Alternative methods of measuring operating speed of electric and traditional bikes in China-implications for travel demand models. J. East. Asia Soc. Transport. Stud. 8, 1424e1436. Cherry, C.R., Weinert, J.X., Xinmiao, Y., 2009. Comparative environmental impacts of electric bikes in China. Transport. Res. Part D: Trans. Environ. 14, 281e290. DeMaio, P., 2009. Bike-sharing: history, impacts, models of provision, and future. J. Pub. Transport. 12, 41e56. Dill, J., Rose, G., 2012. E-Bikes and transportation policy: insights from early adopters. In: Transportation Research Board 91st Annual Meeting, Washington DC. García-Palomares, J.C., Gutiérrez, J., Latorre, M., 2012. Optimizing the location of stations in bike-sharing programs: a GIS approach. Appl. Geogr. 35, 235e246. Gleason, R., Miskimins, L., 2012. Exploring Bicycle Options for Federal Lands: Bike Sharing, Rentals and Employee Fleets. Western Transportation Institute. Gojanovic, B., Welker, J., Iglesias, K., Daucourt, C., Gremion, G., 2011. Electric bicycles as a new active transportation modality to promote health. Med. Sci. Sports Exerc. 43, 2204e2210. Göllei, A., Görbe, P., Magyar, A., 2012. Modeling and optimization of electrical vehicle batteries in complex clean energy systems. J. Clean. Prod. 34, 138e145. Hampshire, R.C., Marla, L., 2012. An analysis of bike sharing usage: explaining trip generation and attraction from observed demand. In: Transportation Research Board 91st Annual Meeting, Washington DC. Ji, S., Cherry, C.R., Bechle, M.J., Wu, Y., Marshall, J.D., 2012. Electric vehicles in China: emissions and health impacts. Environ. Sci. Technol. 46, 2018e2024. Kaltenbrunner, A., Meza, R., Grivolla, J., Codina, J., Banchs, R., 2010. Urban cycles and mobility patterns: exploring and predicting trends in a bicycle-based public transport system. Per. Mob. Comp. 6, 455e466. Kim, D., Shin, H., Im, H., Park, J., 2012. Factors influencing travel behaviors in bikesharing. In: Transportation Research Board 91st Annual Meeting, Washington DC. Krykewycz, G., Puchalsky, C., Rocks, J., Bonnette, B., Jaskiewicz, F., 2010. Defining a primary market and estimating demand for major bicycle-sharing program in Philadelphia, Pennsylvania. Transport. Res. Record: J. Transport. Res. Board 2143, 117e124. Langford, B., Cherry, C., Yoon, T., Worley, S., Smith, D., 2013. North America’s first ebike share: a year of experience. Transport. Res. Record: J. Transport. Res. Board (In Press). Lemire-Elmore, J., 2004. The Energy Cost of Electric and Human-powered Bicycles. Lin, J.-R., Yang, T.-H., 2011. Strategic design of public bicycle sharing systems with service level constraints. Transport. Res. Part E: Log. Transport. Rev. 47, 284e294. Louis, J., Brisswalter, J., Morio, C., Barla, C., Temprado, J.-J., 2012. The electrically assisted bicycle: an alternative way to promote physical activity. Am. J. Phys. Med. Rehab. 91, 931e940. Maurer, L.K., 2012. Feasibility study for a bicycle sharing program in Sacramento, California. In: Transportation Research Board 91st Annual Meeting, Washington DC. Nair, R., Miller-Hooks, E., Hampshire, R.C., Busi c, A., 2012. Large-scale vehicle sharing systems: analysis of Vélib. Int. J. Sust. Transport. 7, 85e106. Parkes, S., Marsden, G., 2012. City bike hire schemes e emerging trends in Europe. In: Transportation Research Board 91st Annual Meeting, Washington DC. Pucher, J., Buehler, R., 2006. Why Canadians cycle more than Americans: a comparative analysis of bicycling trends and policies. Trans. Policy 13, 265e279. Redpath, D.A.G., McIlveen-Wright, D., Kattakayam, T., Hewitt, N.J., Karlowski, J., Bardi, U., 2011. Battery powered electric vehicles charged via solar photovoltaic arrays developed for light agricultural duties in remote hilly areas in the Southern Mediterranean region. J. Clean. Prod. 19, 2034e2048. Rietveld, P., Daniel, V., 2004. Determinants of bicycle use: do municipal policies matter? Transport. Res. Part A: Policy Pract. 38, 531e550. Romero, J.P., Moura, J.L., Ibeas, A., Benavente, J., 2012. Car-bicycle combined model for planning bicycle sharing systems. In: Transportation Research Board 91st Annual Meeting, Washington DC. Rose, G., 2012. E-bikes and urban transportation: emerging issues and unresolved questions. Transportation 39, 81e96. Schroeder, B., Hagen, J., Leve, Z., Penalosa, A., 2009. Bike-sharing goes viral. Sust. Trans., 24e29. Shaheen, S., Cohen, A., Martin, E., 2013. Public bikesharing in North America: early operator understanding and emerging trends. Transport. Res. Record: J. Transport. Res. Board (forthcoming). Shaheen, S., Guzman, S., Zhang, H., 2010. Bikesharing in Europe, the Americas, and Asia. Transport. Res. Record: J. Transport. Res. Board 2143, 159e167. Shang, W., Jiangman, Z., Liang, L., Zheng-yu, D., 2010. Bike-sharing e a new public transportation mode: state of the practice & prospects. In: Emergency Management and Management Sciences (ICEMMS), 2010 IEEE International Conference on, pp. 222e225. Silvester, S., Beella, S.K., van Timmeren, A., Bauer, P., Quist, J., van Dijk, S., 2013. Exploring design scenarios for large-scale implementation of electric vehicles; the Amsterdam Airport Schiphol case. J. Clean. Prod. 48, 211e219. Sperlich, B., Zinner, C., Hébert-Losier, K., Born, D.-P., Holmberg, H.-C., 2012. Biomechanical, cardiorespiratory, metabolic and perceived responses to electrically assisted cycling. Eur. J. Appl. Physiol. 112, 4015e4025.
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024
8
S. Ji et al. / Journal of Cleaner Production xxx (2013) 1e8
Theurel, J., Theurel, A., Lepers, R., 2012. Physiological and cognitive responses when riding an electrically assisted bicycle versus a classical bicycle. Ergonomics 55, 773e781. Tonn, B.E., Schexnayder, S.M., Peretz, J.H., Das, S., Waidley, G., 2003. An assessment of waste issues associated with the production of new, lightweight, fuelefficient vehicles. J. Clean. Prod. 11, 753e765.
Weinert, J., Ma, C., Cherry, C., 2007a. The transition to electric bikes in China: history and key reasons for rapid growth. Transportation 34, 301e318. Weinert, J.X., Burke, A.F., Wei, X., 2007b. Lead-acid and lithium-ion batteries for the Chinese electric bike market and implications on future technology advancement. J. Power Sour. 172, 938e945.
Please cite this article in press as: Ji, S., et al., Electric bike sharing: simulation of user demand and system availability, Journal of Cleaner Production (2013), http://dx.doi.org/10.1016/j.jclepro.2013.09.024