A data-driven dynamic repositioning model in bicycle-sharing systems

doi:10.1016/j.ijpe.2020.107909

International Journal of Production Economics

Volume 231, January 2021, 107909

https://doi.org/10.1016/j.ijpe.2020.107909 Get rights and content

Abstract

The new generation of bicycle-sharing is an O2O (online-to-offline) platform service that enables the users to access the bicycle with a smartphone App. This paper proposes a dynamic repositioning model with predicted demand, where the repositioning time interval is fixed. A data-driven Neural Network (NN) approach is introduced to forecast the bicycle-sharing demand. The repositioning objective function at each time interval is defined to simultaneously minimize the operator cost and penalty cost. In addition to the normal constraints in static repositioning problem, flow conservation, inventory-balance and travel time constraints are taken into account. Due to the non-deterministic polynomial-time hard (NP-hard) nature of this model, a hybrid metaheuristic approach of Adaptive Genetic Algorithm (AGA) and Granular Tabu Search (GTS) algorithm is applied to calculate the solution. Based on predicted demand, the initial repositioning plan is made by AGA statically at the beginning of study horizon, which ensures the global optimization of the first solution. As time goes on, repositioning plan is checked and updated according to the real-usage patterns using GTS algorithm, which has the advantage of high-performance local-search within a short computing time. Numerical analysis is conducted using the real cases. The simulation results reveal that the proposed methodology can effectively model the dynamic repositioning problem in response to real-time bicycle-sharing usage. The proposed methodology can be a value-added tool in enhancing the feasibility and sustainability of bicycle-sharing program.

Introduction

With the increasing recognition of transport sustainability, cycling has become an important alternative mode of public transport for short distance trips. Many cities have successfully set up bicycle-sharing systems to facilitate point-to-point trips, especially for first/last mile trip stages. Bicycle-sharing system (BSS), also known as public bicycle system or bicycle-share system, is a self-service public bicycle rental scheme, which enables a user to borrow a public bicycle at any station and return it to any other station to complete the trip (Zhang and Meng, 2019). The new generation of bicycle-sharing system allows users to find the closest bicycle using their smartphone, scan a Quick Response code and then ride. This one-way usage characteristic tends to cause imbalanced distribution of bicycles across stations over time and space, which can escalate very rapidly during peak hours (Zhang et al., 2019). Take the morning peak as an example, stations with high borrow rate are quick to run empty (e.g. near residential areas, near last-mile transit stops) while stations with a high return rate tend to rapidly become full (e.g. near office buildings, near first-mile transit stops). In this case, users could no longer have bicycles to borrow at the empty stations, nor return bicycles to the full stations. A repositioning vehicle (e.g. a light truck) is used to move bicycles from (near) full stations to (more) empty ones. The repositioning problem has been studied, but it is principally predicated on a static, ‘end-of-day’ method, in which repositioning operation is affected during quiet (night) periods, when the bicycle-sharing demand is negligible (Caggiani et al., 2018). Given the inherently uneven demand-supply (borrow/return) situations at stations, repositioning should be done regularly to address the imbalance of bicycles. Some BSS schemes have introduced intelligent management system to detect the availability of bicycles and docks in each station by means of sensors, predict the forthcoming demand using historical data, and make repositioning decision from the control center in advance to ensure the optimal usage in the system (Zhang et al., 2017). Among these modules, bicycle detection and demand forecasting have been well studied (Lin et al., 2018). Clearly, it is invaluable to develop an optimization method that integrates these modules aiming at timely redistributing the bicycles among stations.

In this paper, a dynamic repositioning model in BSS is developed, where the forthcoming demand is predicted based on a data-driven Neural Network (NN) approach. One case study is conducted to investigate how the proposed model can minimize the vehicle's transportation time and users' waiting time. The contributions of this research are: (1) the proposed method is one of few studies that combines the demand prediction in bicycle reposition, which makes the dynamic operation more accurate; (2) the data-driven NN method has been demonstrated introduced and applied in bike repositioning study; (3) the dynamic repositioning approach could achieve minimum vehicle's transportation time and users' waiting time.

The remainder of the paper is organized as follow. Section 2 summarizes related literature to specify the research problem. Section 3 introduces the data-driven Neural Network approach to predict the bicycle-sharing demand. Section 4 develops the dynamic repositioning model, followed by the solution algorithm in Section 5. Case study is discussed in Section 6, and the conclusion is given in Section 7.

Section snippets

Literature review

The repositioning in BSS relies on an intelligent monitor and scheduling system. Control center first receives real-time usage information from bicycle-sharing stations, then calculates the request for repositioning service, and finally schedules the repositioning vehicle's route as shown in Fig. 1.

However, as the bicycle-sharing usage changes at any time, the repositioning results based on the previous time interval may not meet the requirement of the real-time demand. Therefore, the control

Bicycle-sharing demand prediction

As mentioned earlier, NN method has been widely used in demand prediction in mobility-sharing operations, where the four basic steps of NN include: data collection; network structure definition; network training; and network simulation. The schematic of a typical NN is shown in Fig. 2, which consists of one input layer, one output layer and multiple hidden layers. The network learning process comprises forward computing of data stream and back propagation of error signals. During forward

Vehicle repositioning modelling

The dynamic repositioning model is activated at constant time interval and it uses the forthcoming demand and the system inventory as the input to determine the relocation path and the relocation amount for the repositioning vehicle. Given a bicycle-sharing system, let $N$ represents the set of nodes, $N = N^{n} \cup N^{0}$ ; $N^{n}$ represents the set of bicycle-sharing stations, $N^{n} = {1,2, ..., n}$ , which is indexed by $i, j$ ; $N^{0}$ represents the set of depots, $N^{0} = {0}$ ; $V$ represents the set of repositioning vehicles, which is

Repositioning algorithm

The vehicle repositioning problem is proved to be a type of non-deterministic polynomial-time hard (NP-hard) problem, where solution time increases quickly as the size of the problem grows. Till now, there is no efficient mathematical approach to give an accurate solution to this problem in practice. Herein, metaheuristic technique which can speed up the process of finding a satisfactory solution is widely used in application for solving this NP-hard problem, such as genetic algorithm (GA), ant

Study network

The BSS in Shouguang, China is used as the case study to test the efficiency of the proposed method. This BSS contains 1 depot, 91 stations and 10 repositioning vehicles. The fixed cost for every repositioning vehicle is c₀ = 10 and the variable cost between two nodes is c_v=3.5, which are adopted from Qin's study (2013) based on the real operational setting. The maximum user penalty is c_u = 100, and the large value to avoid unacceptable waiting is set at M = 1000. The distance and travel time

Conclusions

Bicycle-sharing is a convenient and a flexible alternative mode for short-range trips, but the imbalanced demand distribution in bicycles/docks has been a critical issue in almost all BSS, especially during peak hours. This study proposes a dynamic repositioning model in BSS with fixed time interval, where the bicycle-sharing usage is predicted by Neural Network (NN) method. The objective is to minimize the operator cost and minimize user (customer) penalty simultaneously. A hybrid algorithm of

Acknowledgments

The authors would like to thank the anonymous reviewers for their valuable comments and suggestions to improve the quality of the paper.

References (40)

M.M.S. Abdulkader et al.
Vehicle routing problem in omni-channel retailing distribution systems
Int. J. Prod. Econ.
(2018)
L. Caggiani et al.
A modeling framework for the dynamic management of free-floating bike-sharing systems
Transport. Res. C Emerg. Technol.
(2018)
X. Dai et al.
DeepTrend 2.0: a light-weighted multi-scale traffic prediction model using detrending
Transport. Res. C Emerg. Technol.
(2019)
I.A. Forma et al.
A 3-step math heuristic for the static repositioning problem in bike-sharing systems
Transp. Res. Part B Methodol.
(2015)
S.C. Ho et al.
Solving a static repositioning problem in bike-sharing systems using iterated tabu search
Transport. Res. E Logist. Transport. Rev.
(2014)
A. Kaltenbrunner et al.
Urban cycles and mobility patterns: exploring and predicting trends in a bicycle-based public transport system
Pervasive Mob. Comput.
(2010)
M. Kaspi et al.
Regulating vehicle sharing systems through parking reservation policies: analysis and performance bounds
Eur. J. Oper. Res.
(2016)
Y. Li et al.
A multiple type bike repositioning problem
Transp. Res. Part B Methodol.
(2016)
J.R. Lin et al.
A hub location inventory model for bicycle sharing system design: formulation and solution
Comput. Ind. Eng.
(2013)
L. Lin et al.
Predicting station-level hourly demand in a large-scale bike-sharing network: a graph convolutional neural network approach
Transport. Res. C Emerg. Technol.
(2018)

T. Liu et al.

Battery-electric transit vehicle scheduling with optimal number of stationary chargers

Transport. Res. C Emerg. Technol.

(2020)

R. Regue et al.

Proactive vehicle routing with inferred demand to solve the bikesharing rebalancing problem

Transport. Res. E Logist. Transport. Rev.

(2014)

H. Sayarshad et al.

A multi-periodic optimization formulation for bike planning and bike utilization

Appl. Math. Model.

(2012)

C.K. Ting et al.

The selective pickup and delivery problem: formulation and a memetic algorithm

Int. J. Prod. Econ.

(2013)

Y. Wang et al.

Static green repositioning in bike sharing systems with broken bikes

Transport. Res. Transport Environ.

(2018)

C. Xu et al.

The station-free sharing bike demand forecasting with a deep learning approach and large-scale datasets

Transport. Res. C Emerg. Technol.

(2018)

J. Zhang et al.

Bike allocation strategies in a competitive dockless bike sharing market

J. Clean. Prod.

(2019)

J. Zhang et al.

Vehicle routing problem with fuel consumption and carbon emission

Int. J. Prod. Econ.

(2015)

S. Zhang et al.

Electric vehicle routing problem with recharging stations for minimizing energy consumption

Int. J. Prod. Econ.

(2018)

J. Zhang et al.

A dynamic pricing scheme with negative prices in dockless bike sharing systems

Transp. Res. Part B Methodol.

(2019)

Cited by (22)

Teleconsultation dynamic scheduling with a deep reinforcement learning approach
2024, Artificial Intelligence in Medicine
In this study, the start time of teleconsultations is optimized for the clinical departments of class A tertiary hospitals to improve service quality and efficiency. For this purpose, first, a general teleconsultation scheduling model is formulated. In the formulation, the number of services (NS) is one of the objectives because of demand intermittency and service mobility. Demand intermittency means that demand has zero size in several periods. Service mobility means that specialists move between clinical departments and the National Telemedicine Center of China to provide the service. For problem-solving, the general model is converted into a Markov decision process (MDP) by elaborately defining the state, action, and reward. To solve the MDP, deep reinforcement learning (DRL) is applied to overcome the problem of inaccurate transition probability. To reduce the dimensions of the state–action space, a semi-fixed policy is developed and applied to the deep Q network (DQN) to construct an algorithm of the DQN with a semi-fixed policy (DQN-S). For efficient fitting, an early stop strategy is applied in DQN-S training. To verify the effectiveness of the proposed scheduling model and the model solving method DQN-S, scheduling experiments are carried out based on actual data of teleconsultation demand arrivals and service arrangements. The results show that DQN-S can improve the quality and efficiency of teleconsultations by reducing 9%–41% of the demand average waiting time, 3%–42% of the number of services, and 3%–33% of the total cost of services.
A simple agent-based model for planning for bicycling: Simulation of bicyclists' movements in urban environments
2024, Computers, Environment and Urban Systems
Bicycling can improve the sustainability and liveability of cities, many of which desperately require better active transport infrastructure. Urban and transport planners need to examine how improvements in infrastructure change bicyclists' behaviour. With this knowledge, investment in bicycling networks can be more efficient and encourage the use of bicycling for transportation. This study developed a simple Agent-Based Model (ABM) to simulate bicyclists' movements in response to the built environment and road network characteristics in the City of Penrith, in the Greater Sydney Area, Australia. In this case study, the GAMA platform was used to build the ABM and Strava and Riderlog data were used to calibrate and validate the model. The model outputs give insights into bicyclist movements through the road network. The incorporated built environment characteristics include the type of bicycling infrastructure, tree canopy, slope, land use mix, and vehicle traffic. These choice factors also allowed the computation of rider levels of comfort and safety on each trip. Potential refinements of the model include additional bicycling behaviour factors (such as aesthetic preferences), and bicyclists' interactions with each other and other modes of transport.
Efficient inventory routing for Bike-Sharing Systems: A combinatorial reinforcement learning framework
2024, Transportation Research Part E: Logistics and Transportation Review
Bike-sharing systems have become increasingly popular, providing a convenient, cost-effective, and environmentally friendly transportation option for urban commuters on short trips. However, an efficient and sustainable bike-sharing system faces a key challenge to dynamically balancing the supply and demand of bicycles through efficient inventory routing. This paper introduces a comprehensive combinatorial framework that tackles the critical challenges in the bike-sharing system's inventory routing problem. Firstly, we present a novel mathematical model that considers multiple delivery vehicle types and incorporates important factors like dispatch cost, service time, and user satisfaction, all while ensuring fair scheduling. The comprehensiveness of our model makes it highly applicable to real-world scenarios, addressing practical concerns faced by bike-sharing companies. Secondly, we leverage reinforcement learning mechanisms to gather quantitative information on the spatial and temporal patterns of demand and supply. With this data, we construct an effective regression model that accurately predicts station demand. Additionally, we propose an efficient heuristic approach to generate service sequences for delivery vehicle dispatching. Our approach employs a far-sighted strategy-based local iterative search algorithm to construct solutions, coupled with an adaptive exploration algorithm to continually improve solution quality. The proposed solution method is an innovative integration of reinforcement learning, demand prediction, and heuristic-based dispatching, significantly enhancing solution quality and computational efficiency. By bridging the gap between academic research and real-world practice, our framework offers practical and effective solutions for bike-sharing systems. Finally, we validate our proposed framework with extensive experimental results using real-world datasets. Our approach outperforms state-of-the-art algorithms within a short computational time, demonstrating its superiority in terms of solution quality compared to prior literature. Our research opens a new, viable direction for industrial practice, providing valuable insights for decision-makers to optimize bicycle inventory management in a smarter and more efficient way.
Factors affecting the final solution of the bike-sharing rebalancing problem under heuristic algorithms
2023, Computers and Operations Research
The bike-sharing rebalancing problem (BRP) belongs to a class of one-commodity pickup and delivery vehicle route problems (1-PDVRPs) which often must be solved using heuristic algorithms due to the large problem size. However, an open question is which factors affect the quality of the final solutions for the BRP with heuristic algorithms, and how they affect the solution quality. This study proposed six initial solution construction methods and two heuristic algorithms, and applied 32 instances to explore this question in greater depth. The results showed that problem size and the algorithms' searching capacity are the most important factors affecting the quality of final solutions. Furthermore, the influence of the quality and diversity of initial solutions cannot be ignored. The quality of final solutions is negatively correlated with problem size, and positively correlated with the algorithms' searching capacity, and the quality and diversity of initial solutions. The influence of the quality and diversity of initial solutions on the quality of final solutions is positively correlated with problem size. For algorithms with insufficient (powerful) searching capacity, the influence of the quality (diversity) of initial solutions is more significant than that of their diversity (quality). High-quality and rich-diversity initial solutions are very helpful for algorithms to find high-quality final solutions. In addition, the CPU time of heuristic algorithms is generally positively correlated with their searching capacity. These findings have certain universal applicability and reference value for solving the 1-PDVRPs with similar solution structures using heuristic algorithms.
Optimal pricing and design of station-based bike-sharing systems: A microeconomic model
2022, Economics of Transportation
Carefully collected data of nine station-based Bike Sharing Systems (BSS) observed during several years, feed the theoretical formulation of three (aggregated) strategic models representing BSS operation from which the optimal design and pricing is derived. The models include both the operator's costs (investment and operation) and users' costs (time to walk to-from a station, waiting at a station, and time while cycling). The design variables are station spacing, number and capacity of stations, number of bicycles and bike repositioning. Once optimized, the design variables lead to cost functions and optimal pricing. In the first model, a permanent equilibrium without waiting times is assumed. In the second model, waiting at stations (due to a lack of bicycles or docking sites) is introduced in an aggregate form, which results in an increase in the optimal number of bikes and docking sites, making the optimal money price per trip to increase. The third and final model introduces repositioning of bicycles in order to diminish waiting time, making the optimal price grow even further. We obtain an optimal subsidy per trip that grows with the area covered by the BSS, which has implications for its actual implementation in large cities and their spatial and social equity. The optimal pricing scheme is caused by economies of scale due to the reduction in users' access and egress times as the density of stations increases (positive externality) in addition to a fixed operator cost.
A hybrid big-data-based and tolerance-based method to estimate environmental benefits of electric bike sharing
2022, Applied Energy
Citation Excerpt :
At present, the rebalancing of bicycle sharing system includes static rebalancing strategy [44] and dynamic rebalancing strategy [45]. They rebalance the bicycle sharing system by using metaheuristic algorithm [46], deriving service level bounds [47], modifying allocation strategies [48], using stochastic demands models [49], taking Neural Network method [50] or predicting demands [51]. Other creative methods include using robustness for real-time warning [52], rebalancing through pricing strategies [53,54], and rebalancing with the help of deep reinforcement learning frameworks [55].
This research has established a set of methods to use the data of existing docked bike sharing system to pre-estimate the emission reduction effects of docked electric bike sharing system that will be built in the future. It extracted demand through the non-homogeneous Poisson process, and used coarse-grained methods and complex network to estimate total riding distance. Tolerance assumptions were given to accurately estimate the emission reduction effect. Compared with the existing threshold assumption, it can better describe the emission reduction caused by the substitution effect. It is found that the emission reduction caused by the substitution effect after upgrading the docked bike sharing system to the docked electric bike sharing system is 4.19 times of the original. While the reduction in emissions caused by its substitution effect on taxis is three times of the original. This contains a huge emission reduction effect. If 1% of the permanent residents of Nanjing change their commuting vehicles from cars to electric bicycles, the one-way emission reduction will exceed 9.55 tonnes of carbon dioxide. This only requires them to tolerate 5 min in average.

View all citing articles on Scopus

View full text

A data-driven dynamic repositioning model in bicycle-sharing systems

Abstract

Introduction

Section snippets

Literature review

Bicycle-sharing demand prediction

Vehicle repositioning modelling

Repositioning algorithm

Study network

Conclusions

Acknowledgments

Int. J. Prod. Econ.

Transport. Res. C Emerg. Technol.

Transport. Res. C Emerg. Technol.

Transp. Res. Part B Methodol.

Transport. Res. E Logist. Transport. Rev.

Pervasive Mob. Comput.

Eur. J. Oper. Res.

Transp. Res. Part B Methodol.

Comput. Ind. Eng.

Transport. Res. C Emerg. Technol.

Transport. Res. C Emerg. Technol.

Transport. Res. E Logist. Transport. Rev.

Appl. Math. Model.

Int. J. Prod. Econ.

Transport. Res. Transport Environ.

Transport. Res. C Emerg. Technol.

J. Clean. Prod.

Int. J. Prod. Econ.

Int. J. Prod. Econ.

Transp. Res. Part B Methodol.