Space-time demand cube for spatial-temporal coverage optimization model of shared bicycle system: A study using big bike GPS data
Introduction
In recent years, urban transport planners have focused much of their attention on policies promoting the use of bicycles as an alternative to intensive car use (García-Palomares et al., 2012). As a sustainable transport mode, bicycle sharing is increasingly popular and the number of bike-sharing schemes has grown significantly worldwide in recent years (Frade and Ribeiro, 2015). Bike-sharing has several potential benefits not only to society but also to individuals (Mattson and Godavarthy, 2017). The benefits include access to a low-cost public transportation mode, improved health through increased physical activity, flexible mobility, emission reduction, reduced fuel usage, support for multimodal connections, and a reduction in traffic congestion (Shaheen et al., 2010). Bicycle sharing is thus of great significance for improving the quality of urban life, developing sustainable transportation and making better use of urban space.
China's bike-sharing services were launched in 2015 by two startups, OFO and Mobike. This innovative generation of fully dock-less bike-sharing services is based on the prevalence of smartphones and cashless mobile payment and has expanded rapidly since 2016 (Pal and Zhang, 2017). An embedded GPS sensor and communication module are installed on each bike. A customer can easily locate a bike, unlock the bike by scanning its QR code and pay via a smartphone app. With the help of a smartphone application, a bike can be rented and returned anywhere (Yu et al., 2018). According to the statistics of the Ministry of Transport of the People's Republic of China (http://www.mot.gov.cn/) in December 2018, the proportion of bicycle trips in China has doubled to 11.6% since 2015, and the daily use of shared bicycles in China is more than 10 million.
However, with the explosive increase of shared bicycles, some “negative externalities” have emerged. Shared bicycles are free to park at any place and any time on the streets and thus can pose some risk to the public (e.g., pedestrians) and other transportation systems (e.g., vehicular traffic). For instance, unruly parking may lead to the occupation of sidewalks, inconveniencing pedestrians as a result. Forced cleanup may cause serious damage as well as loss and waste of social resources when abandoned or confiscated bikes are piled up like mountains. The above issues point to one of the most important elements in bike-sharing services: the placement of the shared bicycle stations, which is an important way to address these issues (Frade and Ribeiro, 2015).
One important consideration in optimizing the location of bike stations is an accurate assessment of the demand for service. It is a fundamental element in location optimization models (Lin and Yang, 2011; Ren et al., 2016). The classical method for generating the magnitude of demand is based on population or travel surveys. The residential population is most commonly used to represent user demand (Conrowa et al., 2018). These data often rely upon several data collection agencies, such as the Census Bureau. The data (i.e., population) are usually available in an aggregate form at the census-tract or census-block level. However, how individuals are distributed within the enumeration unit is unknown. To obtain a better estimation, García-Palomares et al. (2012) determine the distribution of potential bike trip demand by multiplying the population with coefficients of generation or attraction at the building level. Then, the total number of trips is calculated by adding the number of trips generated and attracted to each building. Travel surveys are also used to provide coefficients of potential demand for bicycle trips throughout a transportation network. For example, the proportions of commuting and recreational bike trips are considered after calculating the origin-destination matrix between traffic zones (Frade and Ribeiro, 2014). Some scholars also consider other factors that influence bike usage when evaluating demand, such as road length, slope, tourist attractions, parks/recreational areas, topography, regional and local transit stations, bicycle-friendly streets, and streets with bicycle lanes. (Gregerson et al., 2010; Krykewycz et al., 2010). To sum up, due to the limitations of conventional data collection methods (e.g., travel surveys with limited numbers of participants for a small number of days or aggregate census data), deriving bike demand using these conventional data also has many limitations. For instance, the potential demand for bicycles has often been considered static and deterministic in previous studies (Lin and Yang, 2011; Park and Sohn, 2017), which means that demand for bicycles in conventional location-based measures is often conceived as temporally invariant and originated from spatially fixed population centers.
However, bike demand distribution varies in space and time in a highly complex manner. Daily cycling demand shows strong temporal and spatial variations. The complexity of urban travel indicates that the spatial structure of bike demand in a city is not fixed because it greatly depends on residents' activity-travel patterns and the urban structure. Similarly, the temporal distribution of bike demand is not static due to the dynamics of individual travel (Chen and Kwan, 2012; Kwan, 2013). In this light, the spatial structure of demand generated using conventional methods may deviate significantly from the structure of the true demand. Such a discrepancy could lead to solutions that may be far from the best. Moreover, current approaches for optimizing the locations of bike stations face various challenges associated with the spatiotemporal dynamics of travel demand, and traditional location models cannot address these temporal issues adequately, especially at specific locations or facilities. This can be a real and major problem in the context of public service provision (Kwan, 2013; Wang et al., 2017).
In recent years, a few scholars have noticed this problem and made some efforts to address them. For instance, Park and Sohn (2017) attempt to make potential demand more precise in the spatial dimension by considering the average floating population in the buffer zones of potential demand sites (e.g., metro stations, shopping centers, parks, and residences). However, they still do not consider the temporal aspect of the potential demand and take into account only limited types of demand zones. Ren et al. (2016) consider individuals' space-time constraints and accessibility to obtain better demand estimation both from the spatial and temporal perspectives. However, the study faces some limitations including the lack of a large real activity-travel dataset.
With unprecedented spatiotemporal coverage as well as richness and granularity, big data have the potential for helping researchers to gain new knowledge about the space-time characteristic of human activities and help generate better solutions for location problems (Tu et al., 2016; Tong and Murray, 2017). The data captured by bike-sharing service providers offer researchers excellent opportunities for tackling this challenge. To use such data to address the problem of locating bicycle stations, the concept of space-time demand cube is presented in this paper to extract fine-grained demand that better reflects the actual demand of bike users. City-wide spatiotemporal variations in bike demand are captured from a large number of GPS records of shared bicycles. Based on the fine-grained demand representation, a spatiotemporal demand coverage method is proposed to optimize the location of shared bicycle stations. Our location model maximizes the space-time demand coverage and minimizes the distance between riders and shared-bicycle stations. The results show that, when compared to population-based models (whether based on fixed or floating population), continuous coverage models that do not take into account the temporal dimension and the Wuhan Public Bicycle System, our proposed model achieves higher coverage of the dynamic spatiotemporal needs of users and can better meet real-world situations.
Section snippets
Study area and data
Wuhan is the capital and largest city of the Chinese province of Hubei, which is also the most populous city in central China. The “China Optics Valley” located in Wuhan is selected as the study area in this paper. It is an area with many leading enterprises and universities and has become a new town of science and technology in the last two decades. Fig. 1 shows the core area of the “China Optics Valley” which covers various types of land use such as universities, high-tech zones
Methodological framework
In this section, we present the basic framework of a space-time maximal covering location model based on the space-time demand cube as shown in Fig. 2. The main procedures of the framework are described as the following four steps. First, demand points in space-time are derived from the bike GPS data that entail a huge amount of users' bike demand. Second, a space-time demand cube set is constructed to represent fine-grained demand, and each demand cube is defined by a set of spatial and
Model evaluation indices
In this paper, the space-time demand coverage rate p, the average distance l and the degree of spatial dispersion s are used as the evaluation indicators of the model, and are expressed as (17), (18), (19) respectively:
In Formula (17), DCi represents the number of demand points covered in the space-time demand cube i, DTi represents the total demand in the space-time demand cube i. In Formula (18), disu represents the distance from covered demand point u
Applicability of the proposed siting method
To obtain a site configuration scheme that better meet the dynamic needs of users, this paper uses big bike data that reflect the characteristics of dynamic demand to measure riding demand. Bike station configuration, however, is only one of the application areas of location models. For other applications of location models, it is also important to consider the spatiotemporal characteristics of the clients of facilities like gas stations (Tu et al., 2016), retail outlets (Kelly, 2017),
Conclusion
Due to data limitations, demand in conventional location optimization models is often treated as temporally invariant or originated from spatially fixed population centers. However, bike demand distribution varies in both space and time in a highly complex manner. Consequently, the spatial structure of demand generated by conventional methods may deviate significantly from the true demand structure, and such a discrepancy may lead to solutions that may be far from the best. In this paper,
Funding
This research was supported by grants from the National Natural Science Foundation of China (grant number 41201385) and the State Key Laboratory of Geo-information Engineering (grant number SKLGIE2017-M-4-1).
References (29)
- et al.
The maximal covering location problem
Pap. Reg. Sci.
(1974) - et al.
Bicycle sharing systems demand
Soc. Behav. Sci.
(2014) - et al.
Bike-sharing stations: A maximal covering location approach
Transp. Res. Part A Policy Pract.
(2015) - et al.
Optimizing the location of stations in bike-shared programs: A GIS approach
Appl. Geogr.
(2012) - et al.
Strategic design of public bicycle sharing systems with service level constraints
Transp. Res. Part E Logistics Transp. Rev.
(2011) - et al.
Bike share in Fargo, North Dakota: Keys to success and factors affecting ridership
Sustain. Cities Soc.
(2017) - et al.
Free-floating bike sharing: Solving real-life large-scale static rebalancing problems
Transp. Res. Part C Emerg. Technol.
(2017) - et al.
An optimization approach for the placement of bicycle-sharing stations to reduce short car trips: An application to the city of Seoul
Transp. Res. A
(2017) - et al.
Optimizing the locations of electric taxi charging stations: A spatial-temporal demand coverage approach
Transp. Res. Part C Emerg. Technol.
(2016) - et al.
Optimizing the configuration of precipitation stations in a space-ground integrated sensor network based on spatial-temporal coverage maximization
J. Hydrol.
(2017)