Next Article in Journal
Centralized Decision Making in an Omnichannel Supply Chain with Stochastic Demand
Previous Article in Journal
Modeling the Antecedents of Green Consumption Values to Promote the Green Attitude
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 15
Issue 17
10.3390/su151713112
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

A Study on Bicycle-Sharing Dispatching Station Site Selection and Planning Based on Multivariate Data

by
Yong Lei
,
Jun Zhang
* and
Zhihua Ren
*
School of Landscape Architecture, Northeast Forestry University, Harbin 150040, China
*
Authors to whom correspondence should be addressed.
Sustainability 2023, 15(17), 13112; https://doi.org/10.3390/su151713112
Submission received: 11 July 2023 / Revised: 29 August 2023 / Accepted: 29 August 2023 / Published: 31 August 2023
(This article belongs to the Section Sustainable Urban and Rural Development)
Browse Figures
Review Reports Versions Notes

Abstract

:
Bicycle-sharing is experiencing rapidly as a low-carbon transport mode of travel, with the advantages of low cost and sustainability. Bicycle-sharing operators use electronic fence parking points to manage bicycle-sharing, but it is too time-consuming and impractical to manage them individually. Therefore, it is necessary to cluster the electronic fence parking points and implement regionalized management. This study uses a Mean-shift clustering algorithm to cluster the electronic fence parking points on Xiamen Island, construct a bicycle-sharing dispatching station, and divide the management area. Singular value decomposition is employed to analyze the travel demand patterns of bicycle-sharing and travel characteristics. In addition, we constructed regression models to explore the relationship between the urban built environment and bicycle-sharing trips during the morning and evening peak hours. The study results show that: (1) The 73 dispatching stations constructed cover 86.72% of the bicycle-sharing data, with a good clustering effect. (2) The travel demand for bicycle-sharing shows multiple patterns and different characteristics in different spatial and temporal distributions, which are influenced by land use. (3) There are spatial and temporal differences in the impact of the urban built environment on bicycle-sharing trips, especially residential and enterprise poi densities have opposite effects on shared bicycle-sharing during morning and evening peak hours. The research results of this paper can serve in the planning of bicycle-sharing dispatch stations and the differentiated management and dispatch of bicycle-sharing, which can reduce operating costs and promote the development of sustainable urban transport.

1. Introduction

Bicycle-sharing plays an essential role in the urban public transport system, promoting sustainable transport development and helping to save energy and reduce emissions. In the urban transport system, bicycle-sharing is the capillary of urban public transport, running through the whole city and meeting the needs of residents in the “last mile” of travel [1]. In short-distance travel, bicycle-sharing can replace traditional fuel vehicles and reduce carbon emissions [2], helping to achieve the goal of “carbon neutrality” in the transport sector [3].
Bicycle-sharing, a sharing economy model, allows users to rent and return bikes at multiple locations in a city without worrying about parking. Unlike traditional bicycles, users pay a small fee for the privilege of using the bike for the duration of the rental period. They are not responsible for the repair and maintenance of the vehicle. The first generation of bicycle-sharing, known as the “white bicycle,” appeared in Amsterdam in 1965 [4]. However, the program had a short lifespan due to frequent theft and vandalism. Subsequently, bicycle-sharing underwent several updates to enhance its security significantly. In 2015, ofo introduced a new generation of bicycle-sharing in China, utilizing GPS technology to track bicycle movements [5]. With the advancement of mobile payment systems, bicycle-sharing rapidly expanded to other countries, such as Singapore and the UK, in 2016. Bicycle-sharing has evolved from initially requiring returns to fixed stations to relying on GPS and personal smartphones, enabling electronic fence parking spots for bicycle returns [6]. In 2016 there were 2 million bicycle-sharing in China, which grew to 23 million in 2017 [7]. In December 2016, bicycle-sharing was introduced in Xiamen, and the number of bicycle-sharing peaked at approximately 468,000 in July 2017. Subsequently, the Xiamen municipal government implemented controls on bicycle sharing. By 2020, the authorities took steps to transition bicycle-sharing from unregulated growth to reorganization, implementing control measures to limit the number of bicycle-sharing to fewer than 150,000 and setting up nearly 15,000 electronic fences [8].
An imbalance in the spatial and temporal distribution of the public’s demand for bicycle sharing may result in no bicycle-sharing being available in certain areas at certain times [9]. To better meet the travel needs of people in different areas and ensure a balance between the supply and demand of shared bikes, bike-sharing operators have invested a lot of effort in managing shared bikes, improving dispatching efficiency, and promoting the sustainability of urban transport [10].
Bicycle-sharing community clustering. Reasonable bicycle-sharing dispatch area planning and site selection can improve the efficiency of bicycle-sharing dispatch. Rodriguez-Valencia [11] used administrative divisions to delineate the study area. However, bike share users may frequently travel between administrative districts, and this top-down administrative division can fragment the independence of bike share travel [12]. The large scope of some administrative divisions is not conducive to bicycle-sharing dispatch and dispatch station planning and site selection. Zhou [13] used bicycle-sharing travel point clustering to delineate the study area through community identification. Caggiani [14] conducted a clustering analysis of the spatio-temporal characteristics of shared bikes, clustering bicycle-sharing with the same travel characteristics in both the temporal and spatial dimensions. This type of method takes the gathering behavior of bicycle-sharing as a reference factor, retains the independence of shared bicycle trips, and the obtained community areas can be used as a basis for bicycle dispatching areas. However, such a bicycle dispatching area ignores the existing dispatching method, and the community bandwidth setting is not considered in conjunction with the existing dispatching method. Therefore, we propose a new method to construct a bicycle-sharing dispatching station using the dispatch distance data of bicycle-sharing and electronic fence data.
The OD matrix is a sparse matrix, and the singular value decomposition can help us extract travel demand’s spatial and temporal patterns from the huge spatio-temporal matrix. This decomposition method can help us understand the demand patterns of bicycle-sharing in cities and the travel demand at different times and locations. Chen [15] used singular value decomposition of traffic data to discover potential traffic patterns and, based on the multidimensional nature of traffic data, recovered the lost traffic data to improve the data quality. Bamaqa [16] decomposes crowd flow data by singular value decomposition for anomaly detection. In traffic accident prediction, Barba [17] proposed a new time series prediction strategy based on singular value decomposition used to obtain the matrix’s low and high-frequency data, which can provide complete and accurate traffic accident prediction results. The spatio-temporal distribution characteristics of bicycle-sharing can be analyzed at a finer scale by utilizing the singular value decomposition of the bicycle-sharing travel matrix. The visualization provides a more direct observation of the spatio-temporal variation in the demand pattern of bicycle-sharing trips. It provides effective dispatching strategies for city managers and bicycle-sharing operators.
Due to the relatively compact urban layout and well-established bicycle lanes in Xiamen, the city provides good conditions for developing Bicycle-sharing. Therefore, selecting Xiamen Island as the study area is important for constructing bicycle-sharing dispatch stations and understanding the impact of the built environment on bicycle-sharing trips, which can help city managers better plan bicycle-sharing dispatch stations and improve the efficiency of the bicycle-sharing system. The study’s results can also serve as a reference for other cities in the field of sustainable transport and promote them to achieve sustainable transport development goals better.
The research objectives of this paper are as follows.
  • To construct a bicycle-sharing dispatching community. Selecting the bicycle-sharing electronic fence as the clustering object, referring to the bicycle-sharing DO point distance (the distance between the endpoint of the last order of the bicycle-sharing and the starting point of the following order), and using the mean-shift algorithm for community clustering to construct dispatching stations and dispatching communities.
  • We perform the decomposition of bicycle-sharing travel demand patterns. The spatiotemporal matrix of bicycle-sharing travel on weekdays is decomposed. Travel demand patterns with large singular values are selected for analysis to observe the influence of the built environment on the travel patterns of bicycle-sharing. The different characteristics of travel demand patterns are used to dispatch bicycle-sharing in time slots to meet users’ travel demands and improve the bicycle-sharing utilization rate.
  • We conducted a regression analysis of bicycle-sharing travel behavior and the built environment. We used the bicycle-sharing dispatching community as the study area and selected the amount of bicycle-sharing trips during the morning and evening peak hours as the study object. By constructing a regression model to assess the impact of the built-up environment on the amount of bicycle-sharing trips, the analysis results can provide targeted guidance and suggestions for bicycle-sharing management schemes and policies to promote the development of sustainable urban transport.
The rest of the paper is organized as follows. The Section 2 proposes a research methodology of bicycle-sharing dispatch community clustering, bicycle-sharing travel demand pattern decomposition, and regression analysis. The Section 3 conducts an empirical study using Xiamen Island as an example. The Section 4 discusses the research results and proposes future research directions. The Section 5 concludes the paper.

2. Materials and Methods

The method framework of this study is illustrated in Figure 1. Multisource data, including bicycle-sharing trip order data, electronic fence parking point data, and built environment data, were collected to support the research on the site selection and planning of bicycle-sharing dispatching stations.

2.1. Research Area

The city of Xiamen is located in the southeastern coastal region of Fujian Province, China, and due to its unique geographical location, the city is divided into two main parts: Xiamen Island and the mainland (Figure 2). We have chosen Xiamen Island as the subject of our study; it is the core area of the city and the main concentration of the city’s population. Xiamen Island consists of two administrative districts: Siming District and Huli District. The Siming district has a more concentrated population than the Huli district (Figure 3). Xiamen has made significant progress in urban public transport and sustainable travel in recent years. As of 2021, Xiamen has three metro lines and 77 metro stations, with a ‘green travel rate’ of 68.2% of its citizens.

2.2. Electronic Fence Parking Spot Clustering

The bicycle-sharing system utilizes electronic fence parking points as the borrowing and returning locations for bicycle-sharing. However, the returning location of the bicycle-sharing may have a certain offset. To address this, we employ KD-Tree to match the bicycle-sharing with the nearest electronic fence parking point (Figure 4). We consider data within a distance of 20 m from the electronic fence parking point as the borrowing and returning data for that specific electronic fence parking point. KD-Tree divides the points in a certain way, forming a binary tree structure [18], and searches for the nearest electronic fence parking point for bicycle sharing. To ensure data validity, we choose to retain electronic fences that are used more than twice a day. We do not consider scheduling problems, as electronic fence parking points with low usage frequencies have a minimal impact on scheduling [19].
Mean-shift clustering is used to cluster the electronic fence parking points as the clustering feature, and the bandwidth is referenced to the distance of the bicycle-sharing DO points (Figure 1). The clustering is performed on the island of Xiamen to form a clustering network for the electronic fence system. The Mean-shift clustering algorithm automatically locates the area with the highest density for clustering [20]. The number of clusters does not need to be specified in advance, and the data can be clustered without prior knowledge of its distribution (Figure 5). Additionally, the computation is faster.

2.3. Singular Value Decomposition

Decomposing the original complex model into several simplified spatiotemporal models better characterizes the intricate spatiotemporal patterns of bicycle-sharing trips. This decomposition approach allows for a more intuitive understanding of bicycle-sharing trips’ changing characteristics and differences in different times and spaces. Singular value decomposition, a commonly used matrix decomposition method, splits a matrix into multiple linearly unrelated components. The technique has been successfully applied to several fields, such as image denoising [21] and motion detection [18]. In the bicycle-sharing domain, the spatiotemporal OD matrix is characteristically decomposed using singular value decomposition to identify the travel patterns of weekday bicycle-sharing and to analyze the spatiotemporal characteristics of different patterns in detail. The singular value measures the importance of each travel demand mode in the spatiotemporal OD matrix of bicycle-sharing trips. The magnitude of the singularity value reflects the amount of basic information contained in the corresponding demand mode.
In order to study the travel demand characteristics of bike-sharing, the spatio-temporal OD matrix of bike-sharing trips from 8 November to 12 November 2021, is subjected to singular value decomposition. It is assumed that this spatio-temporal OD matrix is represented by an m × n OD matrix M (Figure 6). According to the form of singular value decomposition, M can be expressed in the following form:
M = U Σ V T = ∑ h = 1 r δ h u h v h T
where U is the m × m matrix; Σ is the m × n diagonal matrix, his diagonal is the singular value, the size of the singular value reflects the amount of original information contained in the corresponding pattern (retaining the first r singular values); V is the n × n matrix, V T is the transpose matrix of V .
The meaning of the singular value decomposition of the spatio-temporal OD matrix of bicycle-sharing trips is as follows: M represents the set of r types of bicycle-sharing passenger flows, u h represents the temporal distribution of the h-th type of bicycle-sharing passenger flow, δ h represents the singular value of the h-th type of bicycle-sharing passenger flow, and v h T represents the spatial distribution of the h-th type of bicycle-sharing passenger flow.

2.4. Multiple Linear Regression

Based on characteristics such as land use, we extracted factors for the built-up environment of the city (Table 1).
The analysis of the regression model can help us understand the degree of influence of built environment factors on bicycle-sharing cycling behavior [22], investigating the relationship between each built environment factor and the amount of bicycle-sharing use as well as how to promote the use and development of bicycle-sharing by optimizing the built environment.
Due to significant variations in bicycle-sharing usage among different communities, the travel volume of bicycle-sharing does not follow a normal distribution [23]. Additionally, there are significant numerical differences in some built environment factors. We applied a logarithmic transformation to some data to incorporate the data differences more smoothly into the model and obtain more accurate results, reducing the disparities between the data points. After the data transformation (Table 2), the distribution conforms to a normal distribution.

2.4.1. Land Use and Destination Accessibility Data

Land use indicator data is taken from Baidu Maps, including residential poi density, commercial poi density, enterprise poi density, public poi density, green space poi density, building density, and land use function mix. In terms of destination accessibility, the distances between community centers and subway stations, as well as commercial centers, were calculated using Baidu Maps data and ArcGIS Desktop 10.7.

2.4.2. Transport Infrastructure and Socio-Economic Data

Regarding transport infrastructure factors, the density of bus stations was calculated using Baidu Maps data and ArcGIS Desktop 10.7. The density of road intersections was determined using road data from OpenStreetMap and ArcGIS Desktop 10.7. The density of the electronic fence and the idle duration of bicycle-sharing were calculated using Xiamen Big Data Sharing Platform data and ArcGIS Desktop 10.7. Regarding socio-economic factors, the average price of second-hand homes was computed using data from Anjuke. Population data was sourced from WorldPop, and population density was calculated using ArcGIS Desktop 10.7.

2.4.3. Streetscape Data

In the context of streetscape view analysis, we employed the Python-based DeepLabv3 framework to perform semantic segmentation on the street view photos obtained from Baidu Maps [24]. Visual parameters were extracted and computed using the Cityscapes evaluation dataset [25]. We selected the green view index, enclosure density, and rideable road segment as indicators for street view analysis. The calculation method is as follows:
G i = 1 n ∑ i = 1 n P n ( i ∈ N ∗ )
E i = 1 n ∑ i = 1 n P n ( i ∈ N ∗ ) + 1 n ∑ i = 1 n W n ( i ∈ N ∗ )
R i = 1 n ∑ i = 1 n D n ( i ∈ N ∗ )
G i represents the green view index of a streetscape view image, where P n represents the proportion of pixels recognized as trees and other green vegetation, and the sum represents the total number of green vegetation pixels in each street view image. E i corresponds to the enclosure density of the street view image. W n represents the proportion of pixels identified as buildings, walls, and the sum represents the total number of pixels representing buildings, walls, and green vegetation in each street view image. R i represents the rideable road segment of the street view image, while D n represents the proportion of road and sidewalk pixels in the image. Finally, n represents the total number of pixels in each street view image.

3. Results

3.1. Electronic Fence Parking Spot Clustering

Utilizing the electronic fence parking spots for bicycle-sharing dispatch is a primary strategy employed by operators and city administrators [8]. The electronic fence parking spots are distributed throughout various locations in the city, and users can complete their bicycle-sharing trips by returning the bikes to any of these electronic fence parking spots within the city. Returning bicycle-sharing outside the electronic fence parking spots incurs additional dispatch fees, so most citizens park them at the nearest electronic fence parking spot to avoid these extra charges. Electronic fence parking spots help regulate citizens’ bicycle-sharing behavior to some extent. Operators can also sustainably operate the bike-sharing system through this approach. However, due to the complexity and dynamism of citizens’ travel patterns [26], the frequency of using electronic fence parking spots may vary across different areas. Approximately one-third of the electronic fence parking spots had less than 2 usages per day over the past 5 days (Figure 7), and there were nearly 50,000 instances where bike-sharing was not returned to the electronic fence parking spots.
We analyzed the bicycle-sharing travel data and found that the average displacement distance of bike-sharing destination-origin (DO) points was 535.63 m, which can be considered the average distance for bicycle-sharing dispatch. We selected a bandwidth of 500 m for the Mean-shift clustering algorithm and identified 73 communities, which covered 86.72% of the bicycle-sharing origin-destination (OD) data. We established a dispatching station at the center of each community. We used the community boundaries as the dispatching areas to achieve bicycle-sharing dispatching (Figure 8), aiming to optimize the efficiency and sustainability of the bicycle-sharing system.

3.2. Singular Value Decomposition

As a green way to get around, bicycle-sharing promotes low-carbon travel and sustainable development. According to (Figure 9), bicycle-sharing is predominantly distributed in the central areas and industrial parks of Siming and Huli districts. These areas experience a higher frequency of bicycle-sharing usage. The usage volume of bicycle-sharing on various road sections corresponds to the distribution of bicycle-sharing density. Around metro stations, there is a higher flow of bicycle-sharing on the roads and a denser distribution of bicycle-sharing. Conversely, in the outskirts of the city or areas not covered by the metro system, there is lower bicycle-sharing road traffic and a lower density of bicycle-sharing [27].
To comprehensively understand the similarities and differences in the demand patterns of bicycle-sharing travel within the city, we employed the Singular Value Decomposition (SVD) method. By decomposing the spatio-temporal matrix, we analyzed the bicycle-sharing travel data from 8 November to 12 November 2021. According to the scree plot (Figure 10), the largest singular value significantly exceeded the other singular values, and the importance of the singular values decreased rapidly. From the fifth singular value onwards, the magnitude of the value changes became minimal. Therefore, we retained the top 4 singular values, corresponding to demand patterns representing 92.96% of the original information in the spatio-temporal OD matrix.
Based on the singular value decomposition, we identified four distinct travel demand patterns in the spatio-temporal domain of bicycle-sharing travel. A comprehensive view of the fluctuations in demand patterns for all types of shared bicycle travel requires a combination of temporal and spatial vectors to be analyzed. In (Figure 11, Figure 12, Figure 13 and Figure 14), Figure 11a, Figure 12a, Figure 13a and Figure 14a represents the temporal vector of bicycle-sharing travel demand, where positive and negative values indicate the direction of fluctuations in the temporal dimension, and the absolute values represent the magnitude of the fluctuations. Figure 11b, Figure 12b, Figure 13b, Figure 14b illustrates the spatial vector of bicycle-sharing travel demand. Warm colors indicate increased demand; cool colors indicate decreasing demand. And the colors’ darkness or lightness represents the fluctuations’ significance.
Client flow type I represents the daily travel demand for bicycle-sharing (Figure 11). The flow exhibits periodic variations, with similar trends observed each day. Taking Thursday, 11 November, as an example, the data shows two peaks. A significant change at 6 AM indicates the extensive use of bicycle-sharing during rush hour. The data then stabilizes with fluctuations until a second peak occurs at 5 PM, representing the evening rush hour. Regarding spatial distribution, the client flow is primarily concentrated in the central areas of Siming District and Huli District, as well as around Huli Innovation Park and Software Park Phase II. Regarding travel distribution, bicycle-sharing is closely related to the subway network (Figure 4); integrating bicycle-sharing and subways represents a modern way of low-carbon travel [2]. This travel mode helps reduce residents’ reliance on fuel-powered vehicles, lowering carbon emissions, decreasing air pollution, reducing travel costs, and driving the city’s transition to a sustainable transport system [28].
Client flow type II represents the commuting demand between stations and workplaces (Figure 12). The time flow trend for client flow type II is as follows: it increases positively from 7 AM, reaches its maximum positive value at 8 AM, and then rapidly declines, returning to negative values after 9 AM. These trends indicate a significant increase in the demand for bicycle sharing for commuting from stations to workplaces during this period. Commuters may take buses or subways to reach their company’s vicinity and then use bicycle-sharing for the last mile of their journey. In the afternoon, between 5 PM and 6 PM, there is a peak in negative demand for bicycle-sharing commuting, indicating a notable increase in the demand for commuting from workplaces back to the stations. After work, citizens use bicycle-sharing to travel to bus or subway stations, completing the first kilometer of their journey. In terms of spatial distribution, based on the combination of time and spatial unit vectors, the client flow is primarily concentrated in Huli Innovation Park and Software Park Phase II, matching the distribution of subway stations and enterprises.
Client flow type III represents the commuting demand between stations and residential areas (Figure 13). The time flow trend for client flow type III is as follows: it increases positively from 6 AM, reaches its maximum positive value at 7 AM, and then begins to decline. These trends indicate a significant increase in bicycle-sharing demand for commuting from residential areas to stations during this period. Commuters use bicycle-sharing to travel to bus or subway stations, completing the first kilometer of their journey. At 6 PM, there is a peak in negative demand for bicycle-sharing commuting, indicating a notable increase in the demand for commuting from stations back to residential areas during this period. After work, citizens use bicycle-sharing to travel home from bus or subway stations, completing the last kilometer of their journey. In terms of spatial distribution, based on the combination of time and spatial unit vectors, the client flow is primarily concentrated in the central area of Xiamen Island and on both sides of BaiLuZhou Park. This spatial distribution aligns with the distribution of subway stations and residential areas.
Client flow type IV represents the commuting demand during the morning and evening rush hours on working days (Figure 14). The time flow trend for client flow type IV is as follows: it increases positively from 6 AM, reaches its maximum positive value at 7 AM, then rapidly decreases and eventually becomes negative by 8 AM. These trends indicate a significant increase in bicycle-sharing demand for morning rush-hour commuting on working days. Citizens use bicycle sharing to commute during the morning rush hour. A second peak at 6 PM and 7 PM rush hour indicate a noticeable increase in the demand for evening rush hour commuting on working days. Citizens use bicycle sharing to commute during the evening rush hour. In terms of spatial distribution, based on the combination of time and spatial unit vectors, the client flow is primarily concentrated in the central area of Xiamen Island and on both sides of BaiLuZhou Park and around the Huli Innovation Park and Software Park Phase II. This spatial distribution aligns with the distribution of subway stations, residential areas, and businesses.

3.3. Multiple Linear Regression

The average number of bicycle-sharing trips per hour was 3840 for five days from 8 November to 12 November 2021. Looking at the travel patterns decomposed by the singular values, the most significant changes in the number of bicycle-sharing trips were concentrated in the morning and evening peaks, with an average of 7826 trips per hour in the morning peak and 6658 trips per hour in the evening peak (Figure 15) shows the average number of bicycle-sharing trips per hour on weekdays, dividing the time of day into five parts: dawn (3:00–5:59), morning peak (7:00–9:59), daytime (10:00–15:59), evening peak (16:00–19:59), and nighttime (20:00–2:59). Morning peak hour trips are usually higher than evening peak hour trips. The higher number of trips during the morning peak hours can be attributed to individuals in the morning rush hour typically having a clear departure time and needing to ensure they arrive at work on time. They tend to choose bicycle-sharing to complete the last mile to save time and reach their destination quickly. In contrast, during the evening peak hours, people have more flexible commuting options with no clear time constraints, and this flexibility results in relatively low trip volumes during the evening peak hours. As a complement to transport modes, bicycle-sharing meets commuting needs and promotes sustainable transport development.
Construct a regression model (Table 3 and Table 4) with the bicycle-sharing trip volume within the clustered communities as the dependent variable and 18 specific indicators from the built environment index system of 6 categories as independent variables.

3.3.1. Land Use and Destination Accessibility

Regarding land use, we should pay particular attention to the impact of residential and enterprise poi density on bicycle-sharing trip volume within the community (Table 3 and Table 4). A known correlation exists between residential poi density, enterprise poi density, and bicycle-sharing trip volume [29], but this correlation varies over time and space. During the morning peak hours, residential poi density positively correlates with origin trip data while negatively correlates with destination trip data. In contrast, during the evening peak hours, the relationship is reversed, with residential poi density negatively correlated with origin trip data and positively correlated with destination trip data. Compared to residential poi density, the relationship between enterprise poi density and bicycle-sharing trip volume is the opposite. This reflects the commuting demands for shared bicycles between residential and enterprise areas during the morning and evening rush hours [30]. Additionally, mixed land use intensity positively correlates with bicycle-sharing trip volume, indicating that higher land use mix leads to increased bicycle-sharing trips [31]. Adding residential housing near enterprises and improving the land use mix can help mitigate long-distance commuting issues and alleviate road congestion during morning and evening peaks [32].
Regarding destination accessibility (Table 3 and Table 4), the analysis revealed a negative correlation between the distance to the subway station and the commercial district and the number of bicycle-sharing trips. This finding suggests that a shorter distance between the subway station and the commercial district is associated with a higher number of bicycle-sharing trips. Near metro stations, bike-sharing improves the efficiency of travel for the public [23]. The commercial district attracts bicycle-sharing riders as it is a hotspot for various commercial facilities, shopping centers, and entertainment venues. As a complementary mode of transportation, bike-sharing facilitates the last mile of travel, meets the travel needs of citizens, and promotes sustainable transport.

3.3.2. Transport Infrastructure and Socio-Economic

Regarding transport infrastructure (Table 3 and Table 4), the analysis reveals a positive correlation between the density of electronic fences and the number of bicycle-sharing trips. This indicates that areas with a high density of electronic fences are more likely to meet citizens’ demand for borrowing and returning bicycles in bicycle-sharing programs. When there is a high demand for bicycle-sharing trips in an area, city managers will install more electronic fences to manage bicycle-sharing better [23]. Studying the relationship between bicycle-sharing trip volume and bus services provides a way to understand the potential connection between bicycle-sharing and buses. The analysis shows a positive correlation between the density of bus stops and bicycle-sharing use. Increasing the density of bus stops promotes bicycle-sharing use. According to the 2021 Annual Transport Analysis of Major Cities in China, Xiamen citizens’ commuting time is 36.2 min, the average walking distance for people to access the public transport system in Xiamen is approximately 820 m, and the median bicycle-sharing trip is 814 m. Bicycle-sharing has advantages in aiding residents’ commutes. Regarding time benefits, riding a bicycle-sharing to access the public transport system can save time. Compared to walking, riding a bicycle-sharing can reduce commuting time by getting to the bus stop faster [33].
The analysis shows a negative correlation between intersection density and the number of bicycle-sharing trips. A high density of intersections usually means higher traffic volumes and a more complex traffic environment and rules, which may lead to congestion and traffic jams. For cyclists, this may make bicycle-sharing trips less convenient, increasing the risk of traffic accidents and riding insecurity [34], thus reducing the number of bicycle-sharing trips.
There are apparent spatial differences in the idle duration of bicycle sharing (Figure 16). The areas with a shorter idle duration of bicycle-sharing are mainly concentrated in the central area of Xiamen Island, on both sides of BaiLuZhou Park, and around Huli Innovation Park and Software Park II. The areas with longer idle duration for bicycle-sharing are mainly located in the island’s eastern, southern, and northern parts. Relatively small populations characterize these areas; some are still under construction. The long idle duration means that the utilization rate of bicycle-sharing in this area is low, which may be due to three reasons: firstly, the area has a small population, slow economic development, and a lack of large shopping malls or residential areas in the surrounding area, so the use of bicycle-sharing is low, resulting in a large number of bicycle-sharing being idle; secondly, the area is not easily accessible and cannot attract a large number of people; thirdly, the dispatching method of bicycle-sharing in the area is unreasonable, and when people need to ride they cannot find a bicycle-sharing nearby [35], thus reducing their willingness to ride [10]. According to the Xiamen Special Economic Zone Yearbook 2021 [36], the economic level of Siming District is higher than that of Huli District. The length of time between orders in Siming District is smaller than that in Huli District, and the economic level and population concentration in Siming District are higher than that in Huli District. There is a relationship between the distribution of bicycle-sharing idle hours and the city’s economic situation and population density. Operators should improve the circulation rate of shared bicycles according to the actual situation, shorten the length of idle time, reduce the waste of public resources, and promote the sustainable development of the sharing economy.
Regarding socio-economic aspects (Table 3 and Table 4), population density positively correlates with the number of bicycle-sharing trips. Areas with high population density usually have more residential and office buildings with more infrastructure (Figure 17), thus increasing the potential demand for bicycle-sharing. The denser distribution of bicycle-sharing in densely populated areas makes it easier for people to find and use them, increasing the convenience of bicycle-sharing and, thus, the number of bicycle-sharing trips [31].

3.3.3. Streetscape

Regarding streetscape, the positive correlation between the proportion of rideable road segments and the number of bicycle-sharing trips indicates that riders prefer selecting travel routes favorable for cycling, including road segments with wider roads and pedestrian spaces [37]. Wider roads provide cyclists more space, allowing them to maintain a safe distance when sharing the road with other vehicles and enabling cyclists to travel more freely on the road [38].

4. Discussion

Regarding the scheduling and management of bicycle-sharing, the frequency of use of electronic fence parking points and the balance between supply and demand of bicycle-sharing are two key issues. To address these issues, researcher Hui [8] proposed in his study to change the size or distribution of electronic fence parking points according to the intensity of demand for bicycle-sharing. Based on this idea, we suggest adding or removing some electronic fence parking points according to bicycle-sharing usage to meet the public’s demand for borrowing and returning bicycle-sharing [39]. Moreover, by rewarding citizens who park their shared bicycles properly and penalizing those who engage in improper parking, we can encourage citizens to develop a civilized parking habit and improve the order and convenience of the shared bicycle system [40]. By adjusting the layout and quantity of electronic fence parking areas and implementing hybrid incentive programs, we can address the inconvenience of public bicycle borrowing and returning.
In addition, (Figure 18) during the morning and evening rush hours, bicycle-sharing often faces an imbalance between supply and demand [9], especially near residential areas and workplaces. To address this problem, we propose a scheduling community construction method using a Mean-shift clustering algorithm. Compared to other clustering algorithms, K-means clustering [41] needs to predefine the number of clusters, while DBSCAN clustering [42] requires predefining the radius of the neighborhood and the minimum number of data points in the community, which is slower when dealing with large datasets. Mean-shift clustering is more adaptive and can divide the scheduling stations more accurately, with faster computation and better clustering results. This density-based clustering method can adaptively divide an appropriate number of clustered communities by setting different bandwidths [20]. Unlike the previous clustering communities built using bicycle-sharing travel points [13], we borrowed the concept of “central station” [19] and used the Mean-shift clustering algorithm to cluster the electronic fence parking points, and combined it with existing dispatching methods to build a dispatching station and delineate the dispatching range. In this way, we can enhance bicycle-sharing dispatching, improving dispatching efficiency and ensuring a balance between supply and demand, further enhancing bicycle-sharing systems’ sustainability.
Bicycle-sharing is a low-carbon transport mode of travel, reported in Morfeldt’s [43] study as 173 g of carbon dioxide emissions per kilometer from a fuel vehicle. We summarized the mileage traveled by bicycle-sharing over the past 5 days and converted it to the carbon emissions produced by a fuel car traveling the same distance. In total, bicycle-sharing reduced carbon emissions by 83.58 tons. By improving the dispatching efficiency of bicycle-sharing, we can meet the public’s travel needs and further reduce the carbon emissions generated by transportation trips.
Additionally, our research has found differences in the demand for bicycle sharing during morning and evening peak hours. In contrast to previous studies that focused on observing supply-demand relationships in specific hotspot areas [8] or limited regions [9], we identified different travel demands by performing singular value decomposition on the bicycle-sharing demand patterns. Flow type I represents the daily travel demand for bicycle-sharing, flow types II and III represent commuting demands for different purposes during morning and evening peak hours, and flow type IV represents commuting demands specifically during weekdays’ morning and evening peak hours. Based on these different demand patterns, we can implement a time-based scheduling strategy and manage the scheduling and operations of significant hotspot areas within the scheduling communities. By managing bicycle-sharing in these significant hotspot areas at different times, we can reduce resource wastage and better meet bicycle-sharing demands during different periods.
By building a regression model between the built environment and bicycle-sharing trips within the dispatched community, we discovered that the built environment has a certain influence on bicycle-sharing usage. Find the poi density of residences and enterprises had the opposite effect on bicycle-sharing trips in the morning and evening peaks, a situation caused by commuting during peak hours. City administrators can reduce the demand for long-distance commuting by increasing the land use mix. In addition, as the proportion of cyclable road segments increases, more people tend to ride shared bikes as a means of commuting, thereby reducing their reliance on motorized travel [44]. Encouraging people to use bike-sharing for commuting can reduce oil consumption, and combining bike-sharing and public transport reduces travel time [45]. This can promote sustainable transportation and helps alleviate traffic congestion problems [46].
However, there are some limitations to our study. In this paper, we are limited by the size of the bicycle data obtained when studying the characteristics of bicycle-sharing trips and have only learned about trips during weekdays. In the future, if we can get larger-scale data, we can further analyze the bicycle-sharing usage on rest days to compare the similarities and differences between weekday and rest-day trips. This analysis will provide more comprehensive information for managing and scheduling the bicycle-sharing system. Additionally, it is important to consider the impact of other factors on the bicycle-sharing system, such as weather, electric bicycles, road traffic accidents [47], and road congestion conditions [48]. Future research could further explore the effect of these factors on the bicycle-sharing system and propose more comprehensive and holistic scheduling and management strategies.
In conclusion, our research proposes a method for the scheduling and management of bicycle-sharing based on the Mean-shift clustering algorithm and singular value decomposition to build scheduling stations and manage bicycle-sharing in time slots according to their demand, improving the balance of supply and demand and the efficiency of their use, which is conducive to the development of sustainable urban transport.

5. Conclusions

To effectively solve the bicycle-sharing dispatching problem and explore the built environment’s impact on bicycle-sharing travel behavior. This study uses the Mean-shift clustering algorithm to select locations for bicycle-sharing dispatch stations on Xiamen Island to promote sustainable urban transport and shared mobility. And we used singular value decomposition and regression modeling to conduct an in-depth study of the demand patterns and built environment factors for bicycle-sharing trips. The following are the results of our research:
  • Mean-shift clustering algorithm, we selected a bandwidth value of 500 m for electronic fence clustering and eventually identified 73 dispatching stations. These dispatch stations are located around the metro lines and cover 86.72% of the bike-sharing OD data. The effectiveness and rationality of the Mean-shift clustering algorithm in community delineation are demonstrated, providing a useful reference for the planning and layout of bicycle-sharing dispatch stations. Building dispatch stations can reduce the cost of bike-sharing operations, supporting the sustainability of the operating model.
  • The decomposition of bicycle-sharing travel patterns identifies four main passenger flow types. Passenger flow type I represents the daily travel demand for bicycle-sharing, closely related to the metro line. Passenger flow types II and III represent commuting demand between stations and workplaces and between stations and residential areas, respectively, mainly concentrated in the weekday morning and evening peak hours. In addition, passenger flow type IV represents commuting demand during the weekday morning and evening peak hours. Through the changing trends of travel patterns, the time-sharing management of bicycle-sharing in different areas can improve dispatching efficiency and reduce operating costs.
  • By constructing a regression model, we found that built environment factors significantly impact the number of bicycle-sharing trips. Specifically, land use, Destination accessibility, transport infrastructure, and socio-economic and streetscape play a significant role in the number of bicycle-sharing trips. Regarding land use, residential poi density and enterprise poi density have opposite effects on the volume of bicycle-sharing trips at different times of the weekday morning and evening peaks, indicating that the demand for bicycle-sharing trips during peak hours mainly originates from commuting. In addition, the higher the land use mix, the greater the use of bicycle-sharing. Regarding transport infrastructure, the density of electronic fences positively correlates with the number of bicycle-sharing trips, indicating that high-density electronic fences facilitate residents’ borrowing and returning of bicycle-sharing. In addition, the density of bus stops is also positively correlated with the use of shared bikes, and increasing the bus stops density can help promote the use of bicycle-sharing. However, intersection density is negatively correlated with bicycle-sharing trips, implying that excessive density may hurt bicycle-sharing trips. Regarding socio-economic and streetscape, areas with higher population densities are likelier to choose bicycle-sharing trips. In addition, a higher rideable road segment can promote demand for residents to select bicycle-sharing trips.
In conclusion, the research findings demonstrate that the Mean-shift clustering algorithm effectively establishes bicycle-sharing dispatch stations. The decomposition of bicycle-sharing travel patterns allows for a better understanding of the travel patterns of bicycle-sharing, leading to improved dispatching efficiency and reduced operational costs. Additionally, built environment factors have a significant influence on bicycle-sharing travel behavior. These research results provide valuable insights for urban and transportation managers, enabling them to formulate effective bicycle-sharing dispatch strategies and optimize urban transportation layouts. This contributes to enhancing the sustainability of shared mobility and improving the effectiveness of urban traffic management.

Author Contributions

Conceptualization, Y.L. and J.Z.; methodology, Y.L. and J.Z.; software, Y.L.; validation, Y.L., J.Z. and Z.R.; formal analysis, J.Z.; investigation, Y.L.; resources, J.Z.; data curation, Z.R.; writing—original draft preparation, Y.L.; writing—review and editing, J.Z.; visualization, Y.L.; supervision, Z.R.; project administration, Z.R. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Kutela, B.; Teng, H. The influence of campus characteristics, temporal factors, and weather events on campuses-related daily bike-share trips. J. Transp. Geogr. 2019, 78, 160–169. [Google Scholar] [CrossRef]
  2. Saltykova, K.; Ma, X.; Yao, L.; Kong, H. Environmental impact assessment of bike-sharing considering the modal shift from public transit. Transp. Res. Part D Transp. Environ. 2022, 105, 103238. [Google Scholar] [CrossRef]
  3. Li, R.; Li, L.; Wang, Q. The impact of energy efficiency on carbon emissions: Evidence from the transportation sector in Chinese 30 provinces. Sustain. Cities Soc. 2022, 82, 103880. [Google Scholar] [CrossRef]
  4. Anaya-Boig, E.; Douch, J.; Castro, A. The death and life of bike-sharing schemes in Spain: 2003–2018. Transp. Res. Part A Policy Pract. 2021, 149, 227–236. [Google Scholar] [CrossRef]
  5. Guo, Y.; He, S.Y. Built environment effects on the integration of dockless bike-sharing and the metro. Transp. Res. Part D Transp. Environ. 2020, 83, 102335. [Google Scholar] [CrossRef]
  6. Médard de Chardon, C. The contradictions of bike-share benefits, purposes and outcomes. Transp. Res. Part A Policy Pract. 2019, 121, 401–419. [Google Scholar] [CrossRef]
  7. Gu, T.; Kim, I.; Currie, G. To be or not to be dockless: Empirical analysis of dockless bikeshare development in China. Transp. Res. Part A Policy Pract. 2019, 119, 122–147. [Google Scholar] [CrossRef]
  8. Hui, Y.; Xie, Y.; Yu, Q.; Liu, X.; Wang, X. Hotspots Identification and Classification of Dockless Bicycle Sharing Service under Electric Fence Circumstances. J. Adv. Transp. 2022, 2022, 5218254. [Google Scholar] [CrossRef]
  9. Fishman, E. Bikeshare: A Review of Recent Literature. Transp. Rev. 2015, 36, 92–113. [Google Scholar] [CrossRef]
  10. Fukushige, T.; Fitch, D.T.; Handy, S. Can an Incentive-Based approach to rebalancing a Dock-less Bike-share system Work? Evidence from Sacramento, California. Transp. Res. Part A Policy Pract. 2022, 163, 181–194. [Google Scholar] [CrossRef]
  11. Rodriguez-Valencia, A.; Rosas-Satizabal, D.; Unda, R.; Handy, S. The decision to start commuting by bicycle in Bogotá, Colombia: Motivations and influences. Travel. Behav. Soc. 2021, 24, 57–67. [Google Scholar] [CrossRef]
  12. Chen, W.; Chen, X.; Cheng, L.; Liu, X.; Chen, J. Delineating borders of urban activity zones with free-floating bike sharing spatial interaction network. J. Transp. Geogr. 2022, 104, 103442. [Google Scholar] [CrossRef]
  13. Song, J.; Zhang, L.; Qin, Z.; Ramli, M.A. A spatiotemporal dynamic analyses approach for dockless bike-share system. Comput. Environ. Urban. Syst. 2021, 85, 101566. [Google Scholar] [CrossRef]
  14. Caggiani, L.; Ottomanelli, M.; Camporeale, R.; Binetti, M. Spatio-temporal Clustering and Forecasting Method for Free-Floating Bike Sharing Systems. In Advances in Systems Science, Proceedings of the International Conference on Systems Science 2016 (ICSS 2016), Wroclaw, Poland, 7–9 September 2016; Springer International Publishing: Cham, Switzerland, 2017; pp. 244–254. [Google Scholar]
  15. Chen, X.; He, Z.; Wang, J. Spatial-temporal traffic speed patterns discovery and incomplete data recovery via SVD-combined tensor decomposition. Transp. Res. Part. C Emerg. Technol. 2018, 86, 59–77. [Google Scholar] [CrossRef]
  16. Bamaqa, A.; Sedky, M.; Bosakowski, T.; Bakhtiari Bastaki, B.; Alshammari, N.O. SIMCD: SIMulated crowd data for anomaly detection and prediction. Expert. Syst. Appl. 2022, 203, 117475. [Google Scholar] [CrossRef]
  17. Barba, L.; Rodríguez, N. Traffic Accidents Forecasting using Singular Value Decomposition and an Autoregressive Neural Network Based on PSO. Polibits 2015, 51, 33–38. [Google Scholar] [CrossRef]
  18. De, P.; Chatterjee, A.; Rakshit, A. Regularized K-SVD-Based Dictionary Learning Approaches for PIR Sensor-Based Detection of Human Movement Direction. IEEE Sens. J. 2021, 21, 6459–6467. [Google Scholar] [CrossRef]
  19. Huang, J.; Wang, X.; Sun, H. Central Station Based Demand Prediction in a Bike Sharing System. In Proceedings of the 2019 20th IEEE International Conference on Mobile Data Management (MDM), Hong Kong, 10–13 June 2019; pp. 346–348. [Google Scholar]
  20. Pietrzykowski, M. Comparison of mini-models based on various clustering algorithms. Procedia Comput. Sci. 2020, 176, 3563–3570. [Google Scholar] [CrossRef]
  21. Škorić, T.; Pantelić, D.; Jelenković, B.; Bajić, D. Noise reduction in two-photon laser scanned microscopic images by singular value decomposition with copula threshold. Signal Process. 2022, 195, 108486. [Google Scholar] [CrossRef]
  22. Mix, R.; Hurtubia, R.; Raveau, S. Optimal location of bike-sharing stations: A built environment and accessibility approach. Transp. Res. Part A Policy Pract. 2022, 160, 126–142. [Google Scholar] [CrossRef]
  23. Kim, M.; Cho, G.-H. Analysis on bike-share ridership for origin-destination pairs: Effects of public transit route characteristics and land-use patterns. J. Transp. Geogr. 2021, 93, 103047. [Google Scholar] [CrossRef]
  24. Kim, J.H.; Lee, S.; Hipp, J.R.; Ki, D. Decoding urban landscapes: Google street view and measurement sensitivity. Comput. Environ. Urban. Syst. 2021, 88, 101626. [Google Scholar] [CrossRef]
  25. Cordts, M.; Omran, M.; Ramos, S.; Rehfeld, T.; Enzweiler, M.; Benenson, R.; Franke, U.; Roth, S.; Schiele, B. The Cityscapes Dataset for Semantic Urban Scene Understanding. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30 June 2016; pp. 3213–3223. [Google Scholar]
  26. Akova, H.; Hulagu, S.; Celikoglu, H.B. Static bike repositioning problem with heterogeneous distribution characteristics in bike sharing systems. Transp. Res. Procedia 2022, 62, 205–212. [Google Scholar] [CrossRef]
  27. Giuffrida, N.; Pilla, F.; Carroll, P. The social sustainability of cycling: Assessing equity in the accessibility of bike-sharing services. J. Transp. Geogr. 2023, 106, 103490. [Google Scholar] [CrossRef]
  28. Henriksson, M.; Göransson Scalzotto, J. Bike-sharing under pressure: The role of cycling in building circular cycling futures. J. Clean. Prod. 2023, 395, 136368. [Google Scholar] [CrossRef]
  29. Namkung, O.S.; Park, J.; Ko, J. Public bike users’ annual travel distance: Findings from combined data of user survey and annual rental records. Transp. Res. Part A Policy Pract. 2023, 170, 103619. [Google Scholar] [CrossRef]
  30. Goel, R.; Goodman, A.; Aldred, R.; Nakamura, R.; Tatah, L.; Garcia, L.M.T.; Zapata-Diomedi, B.; de Sa, T.H.; Tiwari, G.; de Nazelle, A.; et al. Cycling behaviour in 17 countries across 6 continents: Levels of cycling, who cycles, for what purpose, and how far? Transp. Rev. 2021, 42, 58–81. [Google Scholar] [CrossRef]
  31. Codina, O.; Maciejewska, M.; Nadal, J.; Marquet, O. Built environment bikeability as a predictor of cycling frequency: Lessons from Barcelona. Transp. Res. Interdiscip. Perspect. 2022, 16, 100725. [Google Scholar] [CrossRef]
  32. Glavić, D.; Milenković, M.; Trifunović, A.; Jokanović, I.; Komarica, J. Influence of Dockless Shared E-Scooters on Urban Mobility: WTP and Modal Shift. Sustainability 2023, 15, 9570. [Google Scholar] [CrossRef]
  33. Younes, H.; Zou, Z.; Wu, J.; Baiocchi, G. Comparing the Temporal Determinants of Dockless Scooter-share and Station-based Bike-share in Washington, D.C. Transp. Res. Part A Policy Pract. 2020, 134, 308–320. [Google Scholar] [CrossRef]
  34. Scott-Deeter, L.; Hurwitz, D.; Russo, B.; Smaglik, E.; Kothuri, S. Assessing the impact of three intersection treatments on bicyclist safety using a bicycling simulator. Accid. Anal. Prev. 2023, 179, 106877. [Google Scholar] [CrossRef] [PubMed]
  35. Soltani, A.; Allan, A.; Javadpoor, M.; Lella, J. Space Syntax in Analysing Bicycle Commuting Routes in Inner Metropolitan Adelaide. Sustainability 2022, 14, 3485. [Google Scholar] [CrossRef]
  36. Xiamen Special Economic Zone Yearbook-2021. Available online: http://tjj.xm.gov.cn/tjnj/publish/2021/2021.htm (accessed on 10 July 2023).
  37. Mastora, C.; Paschalidis, E.; Nikiforiadis, A.; Basbas, S. Pedestrian Crossings as a Means of Reducing Conflicts between Cyclists and Pedestrians in Shared Spaces. Sustainability 2023, 15, 9377. [Google Scholar] [CrossRef]
  38. Fitch, D.T.; Carlen, J.; Handy, S.L. What makes bicyclists comfortable? Insights from a visual preference survey of casual and prospective bicyclists. Transp. Res. Part A Policy Pract. 2022, 155, 434–449. [Google Scholar] [CrossRef]
  39. Nikiforiadis, A.; Aifadopoulou, G.; Salanova Grau, J.M.; Boufidis, N. Determining the optimal locations for bike-sharing stations: Methodological approach and application in the city of Thessaloniki, Greece. Transp. Res. Procedia 2021, 52, 557–564. [Google Scholar] [CrossRef]
  40. Choi, S.J.; Jiao, J.; Lee, H.K.; Farahi, A. Combatting the mismatch: Modeling bike-sharing rental and return machine learning classification forecast in Seoul, South Korea. J. Transp. Geogr. 2023, 109, 103587. [Google Scholar] [CrossRef]
  41. Abdulnassar, A.A.; Nair, L.R. Performance analysis of Kmeans with modified initial centroid selection algorithms and developed Kmeans9+ model. Meas. Sens. 2023, 25, 100666. [Google Scholar] [CrossRef]
  42. Chowdhury, S.; Helian, N.; Cordeiro de Amorim, R. Feature weighting in DBSCAN using reverse nearest neighbours. Pattern Recognit. 2023, 137, 109314. [Google Scholar] [CrossRef]
  43. Morfeldt, J.; Davidsson Kurland, S.; Johansson, D.J.A. Carbon footprint impacts of banning cars with internal combustion engines. Transp. Res. Part. D Transp. Environ. 2021, 95, 102807. [Google Scholar] [CrossRef]
  44. Gutiérrez, M.; Hurtubia, R.; Ortúzar, J.d.D. The role of habit and the built environment in the willingness to commute by bicycle. Travel. Behav. Soc. 2020, 20, 62–73. [Google Scholar] [CrossRef]
  45. Böcker, L.; Anderson, E.; Uteng, T.P.; Throndsen, T. Bike sharing use in conjunction to public transport: Exploring spatiotemporal, age and gender dimensions in Oslo, Norway. Transp. Res. Part A Policy Pract. 2020, 138, 389–401. [Google Scholar] [CrossRef]
  46. Sun, S.; Ertz, M. Contribution of bike-sharing to urban resource conservation: The case of free-floating bike-sharing. J. Clean. Prod. 2021, 280, 124416. [Google Scholar] [CrossRef]
  47. Caggiani, L.; Camporeale, R.; Di Bari, D.; Ottomanelli, M. A geofencing-based methodology for speed limit regulation and user safety in e-scooter sharing systems. J. Intell. Transp. Syst. 2023, 1–16. [Google Scholar] [CrossRef]
  48. Eren, E.; Uz, V.E. A review on bike-sharing: The factors affecting bike-sharing demand. Sustain. Cities Soc. 2020, 54, 101882. [Google Scholar] [CrossRef]
Sustainability 15 13112 g001
Figure 1. Methodology framework.
Figure 1. Methodology framework.
Sustainability 15 13112 g001
Sustainability 15 13112 g002
Figure 2. Xiamen City Administrative Divisions.
Figure 2. Xiamen City Administrative Divisions.
Sustainability 15 13112 g002
Sustainability 15 13112 g003
Figure 3. Comprehensive distribution map of Xiamen Island. (a) Population distribution in Xiamen Island; (b) Metro station distribution in Xiamen Island.
Figure 3. Comprehensive distribution map of Xiamen Island. (a) Population distribution in Xiamen Island; (b) Metro station distribution in Xiamen Island.
Sustainability 15 13112 g003
Sustainability 15 13112 g004
Figure 4. KD-Tree schematic. (a) Bicycle-sharing and electronic fence parking spots location distribution; (b) KD-Tree search process; (c) KD-Tree electronic fence search range; (d) KD-Tree electronic fence search range.
Figure 4. KD-Tree schematic. (a) Bicycle-sharing and electronic fence parking spots location distribution; (b) KD-Tree search process; (c) KD-Tree electronic fence search range; (d) KD-Tree electronic fence search range.
Sustainability 15 13112 g004
Sustainability 15 13112 g005
Figure 5. Mean-shift clustering schematic. (a) Calculate the offset mean in the specified area and move the point to the offset mean point; (b) Repeat the above process; (c) Repeat the above process; (d) Satisfy the final condition and end the calculation.
Figure 5. Mean-shift clustering schematic. (a) Calculate the offset mean in the specified area and move the point to the offset mean point; (b) Repeat the above process; (c) Repeat the above process; (d) Satisfy the final condition and end the calculation.
Sustainability 15 13112 g005
Sustainability 15 13112 g006
Figure 6. Singular value decomposition schematic.
Figure 6. Singular value decomposition schematic.
Sustainability 15 13112 g006
Sustainability 15 13112 g007
Figure 7. Low-use electronic fence parking spots.
Figure 7. Low-use electronic fence parking spots.
Sustainability 15 13112 g007
Sustainability 15 13112 g008
Figure 8. Distribution of dispatch stations and dispatch areas.
Figure 8. Distribution of dispatch stations and dispatch areas.
Sustainability 15 13112 g008
Sustainability 15 13112 g009
Figure 9. Distribution of bicycle-sharing use and nuclear density. (a) The distribution of bicycle-sharing nuclear density; (b) The usage volume of bicycle-sharing on various road sections.
Figure 9. Distribution of bicycle-sharing use and nuclear density. (a) The distribution of bicycle-sharing nuclear density; (b) The usage volume of bicycle-sharing on various road sections.
Sustainability 15 13112 g009
Sustainability 15 13112 g010
Figure 10. Singular value distribution of the spatio-temporal matrix of bicycle-sharing trips.
Figure 10. Singular value distribution of the spatio-temporal matrix of bicycle-sharing trips.
Sustainability 15 13112 g010
Sustainability 15 13112 g011
Figure 11. Singular value decomposition results of type I. (a) Time vector distribution; (b) Spatial vector distribution.
Figure 11. Singular value decomposition results of type I. (a) Time vector distribution; (b) Spatial vector distribution.
Sustainability 15 13112 g011
Sustainability 15 13112 g012
Figure 12. Singular value decomposition results of type II. (a) Time vector distribution; (b) Spatial vector distribution.
Figure 12. Singular value decomposition results of type II. (a) Time vector distribution; (b) Spatial vector distribution.
Sustainability 15 13112 g012
Sustainability 15 13112 g013
Figure 13. Singular value decomposition results of type III. (a) Time vector distribution; (b) Spatial vector distribution.
Figure 13. Singular value decomposition results of type III. (a) Time vector distribution; (b) Spatial vector distribution.
Sustainability 15 13112 g013
Sustainability 15 13112 g014
Figure 14. Singular value decomposition results of type IV. (a) Time vector distribution; (b) Spatial vector distribution.
Figure 14. Singular value decomposition results of type IV. (a) Time vector distribution; (b) Spatial vector distribution.
Sustainability 15 13112 g014
Sustainability 15 13112 g015
Figure 15. Mean hourly bicycle-sharing trips by the time of the day.
Figure 15. Mean hourly bicycle-sharing trips by the time of the day.
Sustainability 15 13112 g015
Sustainability 15 13112 g016
Figure 16. Bicycle-sharing idle duration.
Figure 16. Bicycle-sharing idle duration.
Sustainability 15 13112 g016
Sustainability 15 13112 g017
Figure 17. The distribution of mobile phone signal base stations.
Figure 17. The distribution of mobile phone signal base stations.
Sustainability 15 13112 g017
Sustainability 15 13112 g018
Figure 18. Bicycle-sharing Borrow-Return Discrepancy Quantity Comparison Chart for Peak Hour. (a) Bicycle-sharing Borrow-Return Discrepancy Quantity Comparison Chart for Morning Peak Hour; (b) Bicycle-sharing Borrow-Return Discrepancy Quantity Comparison Chart for Evening Peak Hour.
Figure 18. Bicycle-sharing Borrow-Return Discrepancy Quantity Comparison Chart for Peak Hour. (a) Bicycle-sharing Borrow-Return Discrepancy Quantity Comparison Chart for Morning Peak Hour; (b) Bicycle-sharing Borrow-Return Discrepancy Quantity Comparison Chart for Evening Peak Hour.
Sustainability 15 13112 g018
Table 1. Description of explanatory variables for the study.
Table 1. Description of explanatory variables for the study.
Variable Name DescriptionSource
Land useResidential poi density (%)Residential poi density within the communityBaidu Map API
Commercial poi density (%)Commercial poi density within the communityBaidu Map API
Enterprise poi density (%)Enterprise poi density within the communityBaidu Map API
Public poi density (%)Public poi density within the communityBaidu Map API
Green space poi density (%)Green Space poi density within the communityBaidu Map API
Building density (%)Land use mix within the communityBaidu Map API
Land use mix (%)Building density within the communityBaidu Map API
Destination accessibilityDistance to metro station (LN)Logarithm of the distance of the centre of mass within the community from the metro stationBaidu Map API
Distance from shopping area (LN)Logarithm of the distance of the centre of mass within the community from the shopping areaBaidu Map API
Transport infrastructureBus stops density (%)Bus distribution density within the communityBaidu Map API
Intersection density (%)Intersection distribution density within the communityOpenStreetMap
Electronic fence parking point density (%)Electronic fence distribution density within the communityXiamen big data security open platform
Idle duration (LN)Logarithm of the idle duration of bicycle-sharing within the communityXiamen big data security open platform
Socio-economicHouse price (LN)Logarithm of Average Second-hand House Price within the communityAnjuke
Population density (LN)Logarithm of population density within the communityWorldPop
StreetscapeGreen view index (%)Average green view index within the communityBaidu Map API
Enclosure ratio (%)Average enclosure ratio within the communityBaidu Map API
Rideable road segment (%)Average percentage of rideable road segments within the communityBaidu Map API
Table 2. Descriptive statistics of explanatory variables.
Table 2. Descriptive statistics of explanatory variables.
Variable Name MeanStd. DevMinMax
Land useResidential poi density (%)0.0440.0230.0060.113
Commercial poi density (%)0.5980.3980.0961.861
Enterprise poi density (%)0.1690.1250.0110.520
Public poi density (%)0.1270.0890.0140.504
Green space poi density (%)0.0060.0070.0000.033
Building density (%)0.9940.1430.4961.277
Land Use Mix (%)0.2150.0540.0720.330
Destination accessibilityDistance to metro station (LN)6.4650.9863.2708.473
Distance from shopping area (LN)6.4420.8324.3988.393
Transport infrastructureBus density (%)0.0940.0530.0000.285
Intersection density (%)0.0730.0530.0150.308
Electronic fence parking point density (%)0.1220.0770.0110.338
Idle duration (LN)4.6680.2904.1095.410
Socio-economicHouse price (LN)10.8890.2439.90111.384
Population density (LN)2.6211.421−0.2665.627
StreetscapeGreen view index (%)0.2550.0690.1070.385
Enclosure ratio (%)0.4440.0880.2360.629
Rideable road segment (%)0.1500.0140.1150.194
Table 3. Multiple linear regression model (Weekday morning peak).
Table 3. Multiple linear regression model (Weekday morning peak).
Dep. Var: Bicycle-Sharing Weekday Morning Peak OD Count (LN).Origin Dest
βtβt
Land useResidential poi density (%)0.171 **2.925−0.126 *−2.025
Commercial poi density (%)0.205 **3.6650.121 *2.045
Enterprise poi density (%)−0.296 **−4.3730.218 **3.042
Public poi density (%)0.0931.5240.0450.702
Green space poi density (%)0.134 *2.5840.0310.553
Building density (%)0.117 *2.1830.126 *2.207
Land use mix (%)0.0420.705−0.028−0.444
Destination accessibilityDistance to metro station (LN)−0.163 **−3.442−0.133 *−2.647
Distance from shopping area (LN)−0.127 *−2.434−0.165 **−2.987
Transport infrastructureBus stops density (%)0.117 *2.0310.123 *2.022
Intersection density (%)−0.183 **−3.076−0.211 **−3.355
Electronic fence parking point density (%)0.229 **3.2780.478 **6.461
Idle duration (LN)−0.062−1.038−0.091−1.446
Socio-economicHouse price (LN)−0.089−1.528−0.083−1.349
Population density (LN)0.197 **2.9440.163 **2.292
StreetscapeGreen view index (%)0.0640.969−0.108−1.532
Enclosure ratio (%)0.1171.4220.0610.695
Rideable road segment (%)0.147 **2.9190.133 *2.486
N 73 73
F 29.299 25.677
R2 0.695 0.706
Adjusted R2 0.661 0.676
** and * represent 1% and 5% significance levels, respectively.
Table 4. Multiple linear regression model (Weekday evening peak).
Table 4. Multiple linear regression model (Weekday evening peak).
Dep. Var: Bicycle-Sharing Weekday Evening Peak OD Count (LN).Origin Dest
βtβt
Land useResidential poi density (%)−0.210 **−3.4690.118 *2.312
Commercial poi density (%)0.142 *2.4580.098 *2.005
Enterprise poi density (%)0.250 **3.575−0.177 **−2.998
Public poi density (%)0.0781.2410.0821.548
Green space poi density (%)0.0520.9650.099 *2.199
Building density (%)0.136 *2.4530.144 **3.080
Land use mix (%)−0.021−0.3460.0090.172
Destination accessibilityDistance to metro station (LN)−0.202 **−4.128−0.128 **−3.090
Distance from shopping area (LN)−0.198 **−3.680−0.150 **−3.310
Transport infrastructureBus stops density (%)0.175 **2.9510.178 **3.559
Intersection density (%)−0.262 **−4.270−0.161 **−3.107
Electronic fence parking point density (%)0.358 **4.9580.346 **5.692
Idle duration (LN)−0.059−0.966−0.091−1.753
Socio-economicHouse price (LN)−0.097−1.618−0.033−0.654
Population density (LN)0.201 **2.9010.191 **3.271
StreetscapeGreen view index (%)−0.117−1.706−0.027−0.473
Enclosure ratio (%)0.0941.1070.0981.371
Rideable road segment (%)0.114 *2.1890.130 **2.959
N 73 73
F 27.200 39.597
R2 0.701 0.730
Adjusted R2 0.668 0.706
** and * represent 1% and 5% significance levels, respectively.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Lei, Y.; Zhang, J.; Ren, Z. A Study on Bicycle-Sharing Dispatching Station Site Selection and Planning Based on Multivariate Data. Sustainability 2023, 15, 13112. https://doi.org/10.3390/su151713112

AMA Style

Lei Y, Zhang J, Ren Z. A Study on Bicycle-Sharing Dispatching Station Site Selection and Planning Based on Multivariate Data. Sustainability. 2023; 15(17):13112. https://doi.org/10.3390/su151713112

Chicago/Turabian Style

Lei, Yong, Jun Zhang, and Zhihua Ren. 2023. "A Study on Bicycle-Sharing Dispatching Station Site Selection and Planning Based on Multivariate Data" Sustainability 15, no. 17: 13112. https://doi.org/10.3390/su151713112

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop