Professional Documents
Culture Documents
Joint Resource Optimization of Shared Electric Bicycles in Smart Cities
Joint Resource Optimization of Shared Electric Bicycles in Smart Cities
Abstract—As an important part of the green transportation bicycles rely much on those stations since they can provide
of the smart city, the public electric bicycle sharing system has charging services.
been deployed in more and more cities. However, users often RFID [3] or GPS [4] are common equipments of modern
encounter the situation that no bicycle left for rent or no available
dock to return, due to many factors such as dynamic demands bicycles which make it easy to record the history usage
and stochastic usage, which seriously affects the experience of or realtime information of any bicycles. For example, the
users. The state-of-art technology ”borrow anywhere & return number of available bicycles at any station, the borrowing
anyplace” for public bicycles is not suitable for electric bicycles and returning records of any bicycle, or all records of a
since latter ones need docks for batteries recharging. Therefore, given station. Thus, according to the movement pattern, the
the efficient allocation of the stations, bicycles and docks, is
still the key solution to the problem. This paper proposes a future demand could be estimated and then scheduling and
joint optimization of the resources allocation under the condition assignment could be applied prior [5]. These solutions are
of limited number of stations, bicycles and docks, as well as good for dynamic scheduling, however, station localization
the synthetical profiles of the candidate locations. The three and dock allocation must be done ahead to meet the basic
factors are fully considered. Furthermore, part of resources requirement, since all scheduling have to be executed based on
are especially preserved for the future scheduling so that the
proposed scheme can adapt to the dynamics. Users’ behavior is current resources. In practice, due to limited budget, there must
modeled as Poisson distribution for each location. The goal of the be an upper bound for the number of stations and bicycles, as
solution is to optimize the total QoS, not that of a single station. well as number of docks in a station. Therefore, it is vital to
Through the experiments on real data trace, it shows that the allocate resources efficiently. Existing work either considers
proposed resource allocation scheme can achieve better QoS. static deployment, or dynamic scheduling only. For example,
in [6][7][8][9], stationary bicycle stations locations and bicycle
I. I NTRODUCTION allocation are considered. In [2][10][11], dynamic schedulings
are studied. However, although the borrow-return transactions
In favor of advanced information technologies, cities are vary from case to case, from time to time, or from month to
becoming smarter and smarter, not only in management, month, the dynamic is limited in a boundary and based on
but also in services. One of the most important thing that fundamental resources. To maximize the system performance,
people concern is the transportation problem. Although public both stationary allocation and dynamic scheduling must be
transportation systems, such as buses and metro, have been considered simultaneously.
serving people for more than one hundred years, they still fail In this paper, we propose to implement a joint optimization
to provide individual and on demand services. Taxis are good of three basic resources of electric bicycle system, i.e., station
supplement for private services, however, their profits force deployment, dock assignment and total number of electric
them expect long distance transport. Public [1] or shared [2] bicycles. To be distinguished from existing work, in the
bicycles are emerging in many countries which are promising proposed method, part of resources are reserved for future
means to solve the ”last mile” transportation problem, but there dynamic scheduling. The portion is carefully chosen to balance
still exists some demands waiting to be fulfilled. For example, the basic requirements and dynamic ones. Three steps are
transportation with moderate distance, is too painful for a adopted in the scheme. The first step is to select the locations
bicycle ride (in a bad weather condition) or very disappointing of stations according to user gathering points [12]. The second
for a car driving (in a bad traffic jam). Electric Bicycles step is to assign major part of docks and the total number of
have been considered as an important supplement to current electric bicycles. The last step is to assign dynamic docks
public bicycle systems. They are powered by electricity so and bicycles according to flow of bicycles between stations.
that they need less labour and run further and faster. The The optimization problem is NP hard and has no polynomial
management and scheduling of public bicycles has been an solution. Therefore, to reduce the computing cost, we propose
hot topic for a while, which aims to improve the efficiency a greedy algorithm to approximate the optimal solution. The
and user experience. The electric bicycle system has the same future work includes working on a more precisely solution
challenges as the traditional bicycles. However, compared with based on the prediction of user mobility and city open data
shared bicycles, which do not need docking stations, electric as well as incentive strategies to encourage user to borrow or
1139
slot while Ti (out) stands for the number of borrowed ones. E
Based on the statistic of borrowing and returning information, A
we have algorithms to schedule auxiliary docks and electric
bicycles. GP 1 C
D(300m) G GP 2
IV. J OINT OPTIMIZATION
The problem proposed above is a NP hard problem, we
try to bring forward a near optimal solution in this paper. B
There are three steps for the algorithm. First, we assume there
is no limitation of the number of available docking stations. GP 3
F
Since each GP will have a docking station, we assign electric
bicycles to each station and sort them in descending order. D
1140
m
λxj −λ
Pj (Vj (OU T ) = x) = e , x = 0, 1, ... (3) L(k) = (kj , whererj = 0) (7)
x!
j=1
It is easy to calculate the value of λ, for example, in Fig. 2, These bicycles will be used as auxiliary resources, as shown
for GP3 , parameter λ for borrowing is 2 which implies that in the following subsection.
two users having valid visit will leave GP3 .
In this paper, we hope there is a high rate of successful Algorithm 1 Bicycle allocation algorithm
service at each station. The quality of service when assigning Input: P (λj ), K
kj electric bicycles at Gj during the time slot T is SerQj (kj ) Output: k
1: initiate
k = 0, 0, ..., 0
kj
2: FOR i = 1 to K DO
SerQj (kj ) = P (i, λj ) (4)
3: FOR j = 1 to m DO
i=1
4: compute ΔSerQj (kj ) = P (X = (kj + 1)) − P (X =
Then we have the globe optimization target function and kj )
limitations formulated as below 5: END FOR
6: Find maximal ΔSerQj (kj )
m kj
7: kj = kj + 1
M aximize F (k) = (rj × P (i, λj )) 8: End FOR
j=1 i=1
9: Return k = {k1 , k2 , ..., km }
m
s.t. kj = K, (5)
j=1 C. Estimation of docks at each station
m
Besides borrowing demand, users also need to return their
rj = R, rj ∈ {0, 1} electric bicycles. For electric bicycles, the dock is not only
j=1
acting as a lock but also as a charger. User must return
The above optimization problem is NP hard, so we propose it properly can it be charged to serve the next customer.
to use greedy algorithm to finish the second step. First we If there are not adequate docks, user cannot return their
assume the resources are adequate, therefore each GP can have bicycles. They might need to wait or return bicycles to further
a station, then we can remove the second constraint, we have stations, which brings great inconvenient. The simple way is
kj
m
to assign the same number of docks as the bicycles at the
beginning. However, when the system is in operation, it is not
M aximize F (k) = P (i, λj )
necessary that borrowing and returning remain balance so that
j=1 i=1
m
(6) the estimation of docks must be considered separately.
s.t. kj = K According to Def. 2, in a time slot T , the maximal demand
j=1 for docks at GP Gj is
We propose Alg. 1 to solve it. The main idea is at each
M axQj = M axUj = Vj (8)
round, we add one electric bicycle. We calculate service
quality of each station SerQj (kj ) when the new bicycle is
assigned to it. Finally, we chose the station with maximal Also, not all users that visited GP will return bicycles,
increasing service quality to assign the bicycle. Notice that, similarly, we record the returning status at GP Gj which abides
in step 5, if the current increment is the same, we compare by poisson distribution as well. ηj is used to indicate the pa-
the future increment, and chose the one with maximal value. rameter. To reserver some docks for future scheduling, we set
Because only the local situation is considered, after all bicycles aside L(s) docks where L(s) = L(k) (each reserved bicycle
have been assigned, the total increment is close to the optimal is corresponding to a dock, it is a conservative solution). From
solution. above section, we formulate the problem as
Now we introduce the resource limitation that the total
number of available docking station is only R. Vector kj is m
sj
sorted in descending order and the first R elements will be M aximize F (s) = ( P (i, ηj ), whererj = 1)
j=1 i=1
assigned to stations. If ki and kj are equal, then sort them
m
according to λi and λj . If they are still equal, then sort (9)
s.t. sj = S − L(s),
them by Vi and Vj . After sorting, the number of bicycles is
j=1
corresponding to Eq.(6). When R is less than the number of
GPs, some GPs without stations are still assigned bicycles sj ≥ kj
in the first step. In that case, these additional bicycles are The problem can be solved using similar solution as men-
represented by L(k), and tioned above.
1141
D. Assignment of auxiliary docks and bicycles
To deal with the higher demand during peak time, in this pa-
per, we propose to use auxiliary dock and bicycles. In another
word, to reserve some resource and conduct prior assignment
at certain stations that may encounter peak demands. With
additional reserved resources, this can help those dynamic
algorithms work more efficiently.
If bicycles resources are not enough to meet the requiremen-
t, i.e., L(s) = L(k) = 0, it is not possible for any reservation. (a) Station 285 (b) Station 499
If L(s) = L(k) > 0, we formulate the problem as Fig. 3. One day record for borrowing and returning at two stations
m L(k)
j
M aximize ( P (i, (λj − ηj )), whererj = 1 λj > ηj )
ratio. We assume all data from Citybike is the real demand for
j=1 i=1
any station. Based on that, we can test our methods through
m
reducing available resources. The initial number of dock is
s.t. L(k)j = L(k) 120% of the total electric bicycles.
j=1
(10)
According to Alg.1, the near optimal solution can be
achieved either. The calculation of auxiliary docks are little
bit complex than auxiliary bicycles. First, it must satisfy those
stations that have been assigned with auxiliary bicycles. Then
the additional docks can be assigned to other stations. There
are two steps.
(1) Assign docks to auxiliary bicycles, i.e., for any GP Gj ,
assign L(k)j − (sj − kj ) docks to it;
(2) Use following equation to get optimal allocation for the
rest of auxiliary docks:
m L(s)
j
M aximize ( P (i, (ηj − λj )), whererj = 1 ηj > λ j )
j=1 i=1
Fig. 4. Service quality v.s. total number of electric bicycles.
m
s.t. L(s)j = L(ŝ)
j=1
(11)
In summary, our scheme firstly ensures the basic require-
ment in the whole area with the constraints of resources. Then,
if additional resources are available after step one, we will
allocate these spare docks and bicycles using similar algorithm
to achieve future scheduling performance.
V. S IMULATION AND EXPERIMENTS
We use the trace data from New York public shared bicycles
system [19]. Fig. 3 shows two stations in Citybike (ID 285 and
499). It contains 24 hours borrowing and returning data. Two
stations have similar pattern. During hour 0 to 8, there are less
borrowing and returning activities, the peak time is afternoon
and before nightfall. During borrowing peak, less bicycles
are returned (station 499 at 4pm), or less borrowing during Fig. 5. Service quality v.s. total number of docks.
returning peak (station 285 at 6pm). Thus, the station must
have sufficient bicycles or docks to provide service, rather As shown in Fig. 4, the y-axis represents the total number of
than rely on the balance of borrowing and returning. electric bicycles in the system. If the resource is adequate, the
We evaluate the system performance from four aspects, proposed algorithm can satisfy the demand (borrow). When
namely, average service quality (borrow), average service reduce the available bicycles in the system, our algorithm
quality (return), bicycle utilization ratio, and dock utilization still can achieve near 100% service quality, until the available
1142
bicycles drops to 80%. Via average utilization, we could see [10] J. Zhang and P. S. Yu, “Trip route planning for bicycle-sharing systems,”
in CIC. IEEE Computer Society, 2016, pp. 381–390.
that many bicycles are in idle state. By applying the proposed [11] T. Benarbia, K. Labadi, A. M. Darcherif, and M. Chayet, “Modelling
method, it can greatly improve the utilization, which will and control of self-service public bicycle systems by using petri nets,”
bring higher service quality with less resource requirement. Int. J. of Modelling, vol. 17, pp. 173–194, Dec. 17 2014.
[12] B. Fan, S. Leng, K. Yang, and Q. Liu, “GPS: A method for data sharing
However, 100% is not possible to achieve. in mobile social networks,” in Networking. IEEE, 2014, pp. 1–9.
Fig. 5 shows the average service quality for returning and [13] Y. Lin and W. Dou, “Research on demand prediction of urban bicycle
the utilization of docks. Similar to bicycles, through global sharing based on network model,” Application Research of Computers,
vol. 34, no. 9, pp. 1–5, 2016.
optimization, reduce redundant docks, and ensure the quality [14] H. Dong, M. Wu, D. Liu, and N. Chen, “Estimation method of natural
of services. demand of urban public bicycle system,” Journal of Zhejiang University
(Engineering Science), vol. 50, no. 2, pp. 265–270, 2016.
VI. C ONCLUSION [15] Y. ranzi Xu and J. Shen, “Demand analysis of public bicycle system
based on circular distribution method and time series model,” Industrial
In this paper, a joint optimization method is proposed to Engineering Journal, vol. 17, no. 2, pp. 55–63, 2014.
improve the quality of service for the emerging new trans- [16] S. Zhang, Y. Ji, S. Dong, and J. Zhou, “Gravitational field modeling for
distribution characteristics of public bicycle rental duration,” Journal
portation means of electric bicycles. The major constraints of Transportation Systems Engineering and Information Technology,
are the number of docking stations, the total number of vol. 16, no. 2, pp. 200–205, 2016.
electric bicycles and the number of docks. We build the [17] L. Chen, D. Zhang, G. Pan, X. Ma, D. Yang, K. Kushlev, W. Zhang, and
S. Li, “Bike sharing station placement leveraging heterogeneous urban
possibility model of docking station allocation, and propose open data,” in UbiComp, K. Mase, M. Langheinrich, D. Gatica-Perez,
a serial of target function. Furthermore, auxiliary resources H. Gellersen, T. Choudhury, and K. Yatani, Eds. ACM, 2015, pp.
such as bicycles and docks are considered for future dynamic 571–575.
[18] E. Côme and L. Oukhellou, “Model-based count series clustering for
scheduling, and additional assignment is conducted. bike sharing system usage mining: A case study with the Vélib’ system
The simulation is based on the public bicycles of New York of paris,” ACM TIST, vol. 5, no. 3, pp. 39:1–39:21, 2014.
city. The results show that our method can improve the quality [19] Citibike, “Citi bike trip histories,” https://www.citibikenyc.com/system-
data, 2015.
of service efficiently. Namely, our method needs less resource [20] Metrobike.LLC, “The bike-sharing world - year end data,” http://bike-
while provide same quality. The future work includes further sharing.blogspot.com/2016 01 01 archive.html, 2015.
optimization on reservation of bicycles and docks which will [21] D. Zhou, “The development and trend of public bicycles in china,” China
Bicycle, vol. 2012, no. 10, pp. 52–55, 2012.
better fit the dynamic scheduling. [22] E. Co.Ltd, “The first step in china – launching of the smart shared
bicycles,” http://www.edaibu.net, 2016.
ACKNOWLEDGMENT
This work is supported by the Natural Science Foun-
dation of China (61601157), Chinese Scholarship Council
(201208330096).
R EFERENCES
[1] X. Xie, J. Qiu, and M. Tang, “Demand prediction of public bicycle
rental station based on elman neural network,” Computer Engineering
and Applications, no. 8, pp. 1–5, 2016.
[2] L. Chen, D. Zhang, L. Wang, D. Yang, X. Ma, S. Li, Z. Wu, G. Pan,
T. M. T. Nguyen, and J. Jakubowicz, “Dynamic cluster-based over-
demand prediction in bike sharing systems,” in UbiComp. ACM, 2016,
pp. 841–852.
[3] K.-Y. Lin, M.-W. Hsu, and S.-R. Liou, “Bicycle management systems in
anti-theft, certification, and race by using rfid,” in Proceedings of 2011
Cross Strait Quad-Regional Radio Science and Wireless Technology
Conference, 2011, pp. 1054–1057.
[4] L. Chen, D. Zhang, G. Pan, L. Wang, X. Ma, C. C. 0004, and
S. Li, “Container throughput estimation leveraging ship gps traces and
open data,” in The 2014 ACM Conference on Ubiquitous Computing,
UbiComp ’14, Seattle, WA, USA, September 13-17, 2014, A. J. Brush,
A. Friday, J. A. Kientz, J. Scott, and J. Song, Eds. ACM, 2014, pp.
847–851.
[5] Y. Li, Y. Zheng, H. Zhang, and L. Chen, “Traffic prediction in a bike-
sharing system,” in SIGSPATIAL/GIS. ACM, 2015, pp. 33:1–33:10.
[6] L. Li, H. Chen, and X. Sun, “Bike rental station deployment planning
in wuhan,” Urban Transport of China, vol. 7, no. 4, pp. 38–44, 2009.
[7] L. He, D. Chen, X. Li, and J. Lu, “An optimization model of the layout of
public bike rental stations,” Journal of Wuhan University of Technology,
vol. 36, no. 1, pp. 38–44, 2012.
[8] Z. Zhang, P. Guo, and X. Li, “Simulation analysis and optimization
of deployment number of bikes in pub-lic bicycle sharing systems,”
Computer Engineering and Applications, no. 6, pp. 1–6, 2016.
[9] P. Vogel and D. C. Mattfeld, “Strategic and operational planning of bike-
sharing systems by data mining - A case study,” in ICCL, ser. Lecture
Notes in Computer Science, vol. 6971. Springer, 2011, pp. 127–141.
1143