A cluster-based shared bicycle spatiotemporal layout determination method

Abstract

This invention discloses a clustering-based method for determining the spatiotemporal layout of shared bicycles, belonging to the field of intelligent transportation technology. First, the invention systematically analyzes the current status of public bicycles to understand the characteristics of the existing system, and establishes buffer zones using ArcMap software. Second, it employs fuzzy C-means clustering analysis to cluster shared bicycle stations based on differences in daily usage time, obtaining the activity characteristics of various shared bicycle clusters. Finally, based on Points of Interest (POIs), a binary Logit model is used for regression analysis to obtain the locational characteristics of shared bicycle parking layouts. This invention provides an effective basis for travelers to make better travel decisions, and enables the rational allocation of public bicycle demand, thus improving the traffic environment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation technology, specifically relating to a clustering-based method for determining the spatiotemporal layout of shared bicycles. Background Technology

[0002] In recent years, the rapid increase in the number of private cars has brought about numerous problems such as traffic congestion, making the development of public transportation a desirable solution. Guiding drivers from solely relying on private cars to a combination of car and public transportation, and even ultimately to a stage where they rely entirely on public transportation, has become a key focus. Meanwhile, to maximize the advantages of various modes of transportation and minimize overall costs, or because a single mode of transportation is insufficient, travelers often opt for combined travel involving transfers between multiple modes. Among these options, shared bicycles offer a relatively reasonable solution to the "first mile" and "last mile" problems.

[0003] Currently, China's public bicycle industry is experiencing rapid growth, with public bicycle sharing systems (BSS) expanding across the country. However, the challenges in design, planning, management, and operation are also intensifying. A more comprehensive understanding of the supply and demand patterns and the spatiotemporal distribution of usage patterns within the BSS system is needed, requiring the development of a series of models to improve our understanding of the system. In conventional models, BSS station entry and exit analyses are modeled separately, neglecting the close relationship between the entry and exit (arrival and departure) processes at each BSS station. Furthermore, current models often overlook the interactions between neighboring stations, leading to estimation biases, inconsistencies, or inefficiencies. Summary of the Invention

[0004] This invention addresses the shortcomings of existing models and analysis techniques by proposing a clustering-based method for determining the spatiotemporal layout of shared bicycles. The public bicycle parking characteristic profile based on subway and bus stations is a core issue in the public bicycle system.

[0005] This invention specifically includes the following steps:

[0006] 1. Data Analysis and Processing of Shared Bicycle Trips

[0007] A systematic analysis of the current state of public bicycles is conducted to understand the characteristics of the existing public bicycle system, and a spatiotemporal distribution analysis is performed on the acquired public bicycle travel data. Using the time and latitude / longitude of shared bicycles entering and leaving stations as basic data, transportation zones are established centered on subway and bus stations.

[0008] Before processing, the analysis conditions are first determined: the number of valid subway stations, the number of valid bus stations, a certain number of random selections of ordinary BSS stations, and the morning and evening peak hours.

[0009] Before running the next fuzzy clustering algorithm model, a buffer zone was established using Arcmap software (a study area with a radius of 300 meters centered on each station). The analysis time period was divided into two categories: weekdays and weekends. Spatial cross-analysis was performed on the shared bicycle flow entering and leaving the station and the buffer zone range of the two types of stations to obtain all data streams of entering and leaving the station, as well as the data streams of entering and leaving the station during the morning peak and evening peak.

[0010] Data on all shared bicycles entering and exiting BSS stations was divided into eight groups: weekday exit data around subway stations, weekday entry data around subway stations, weekday exit data around bus stops, weekday entry data around bus stops, weekend exit data around subway stations, weekend entry data around subway stations, weekend exit data around bus stops, and weekend entry data around bus stops. The entry and exit data were then integrated into three sets of overall analytical data.

[0011] Based on all the obtained data, the average traffic flow every ten minutes for all BSS (Bike Service Station) stations' shared bike entry and exit data was calculated to facilitate analysis and clustering. In calculating the average traffic flow every ten minutes for all shared bikes entering and exiting stations, the traffic flow from 00:30 to 04:30 was considered negligible, and 20 hours were selected as the effective time period within a day. Based on the average traffic flow every ten minutes for all BSS stations, the ratio of the average traffic flow every ten minutes during the morning and evening peak hours was calculated, denoted as the morning peak entry / exit ratio and the evening peak entry / exit ratio. The corresponding traffic flow ratios for the morning and evening peak periods were also calculated.

[0012] 2. Use fuzzy C-means clustering analysis to obtain the activity characteristics of each cluster.

[0013] A soft clustering technique called fuzzy C-MEANS clustering algorithm is used to cluster shared bicycle stations based on differences in daily usage time. By continuously optimizing the objective function, the membership degree of each sample point to each cluster center is obtained, and the highest membership degree is selected as the classification criterion, thereby achieving the goal of automatically classifying the sample data. The algorithm aims to solve the following objective function:

[0014]

[0015]

[0016] Where c is the predetermined number of clusters, m is the fuzziness, and fuzziness is a parameter that determines the impact of outliers (data objects that do not belong to a specific cluster) on the cluster centroid values. N is the number of discrete clusters. ik This represents the membership value of the cluster from the i-th site to the k-th site. For a given site, the sum of its membership values ​​with respect to all other clusters must be 1. i ,x k L(x) represents the value of the i-th object and the value of the k-th cluster centroid, respectively. i ,x k The membership value is a measure of the dissimilarity between the i-th object and the k-th cluster centroid. The sum of squared residuals is used as the measure of anisotropy. The optimal condition is achieved when the membership value satisfies the following condition:

[0017]

[0018] In the process of fuzzy clustering, the time period is divided into weekdays and weekends, and the shared bicycle flow entering and leaving the BSS station is considered separately. After determining the predetermined number of clusters, three indicators are selected as the dissimilarity measurement objects: the average flow rate every ten minutes for all entering and leaving the station, the ratio of the morning peak to the evening peak for entering and leaving the station, and the squared error is used as the dissimilarity measurement value. Modeling analysis is performed on the eight groups.

[0019] The silhouette coefficient method was selected as the main basis for determining the number of clusters in each group, and the sum of squared errors (SSE) was combined to determine the number of clusters in each group of data.

[0020] The core metric of this method is the silhouette coefficient, which is the metric for a given sample point X. i The contour coefficient is defined as follows:

[0021]

[0022] Where a is X i The average distance to other samples in the same cluster is called cohesion, and b is the distance of X. i The average distance to all samples in the nearest cluster is called the dissociation. The nearest cluster is defined as follows:

[0023]

[0024] Where p is a certain cluster C k The samples in. That is, using X. i After using the average distance to all samples in a cluster as a measure of the distance from a point to that cluster, select the distance from X. i The most recent cluster is designated as the nearest cluster.

[0025] The average silhouette coefficient is obtained by calculating the silhouette coefficient of all samples according to the formula and then averaging them, with a value range of [-1, 1]. The closer the samples within a cluster and the farther the samples between clusters, the larger the average silhouette coefficient and the better the clustering effect. Therefore, k, which has the largest average silhouette coefficient, is the optimal number of clusters. Statistical analysis was performed on each station based on the average number of shared bicycles entering / exiting the station every ten minutes throughout the day, the ratio of entering / exiting the station during the morning peak, and the ratio of entering / exiting the station during the evening peak. The C-MEANS clustering algorithm based on minimizing the weighted squared error function was applied to divide the stations into clusters based on the daily flow and time of shared bicycles. The stations were then classified according to the main cluster value with the highest membership degree, resulting in the activity characteristics of various shared bicycle clusters.

[0026] 3. Location Feature Regression Analysis of Shared Bicycles Based on POI

[0027] To assess the relationship between station type and corresponding locational characteristics, it is necessary to acquire relevant geographic information data within a 300-meter radius of subway stations and bus stations. Using the data cloud platform of Gaode Maps Open Platform and a data acquisition model built with Python software, over ten geographic data points (POIs, representing a specific feature area in a geographic information system) within a 300-meter radius of each station were obtained, including schools, residential areas, restaurants, sports venues, shopping centers, and companies. A binary Logit model was used for regression analysis to obtain the locational characteristics of shared bicycle parking layouts.

[0028] The binary logistic regression model can reveal the interactions and communication structures between spatial elements. Based on the continuous membership values ​​generated by the C-MEANS clustering method, the binary logistic regression model can be used to explore the determinants of cluster membership while controlling for spatial autocorrelation between sites. The model formula is established as follows:

[0029]

[0030] In the above formula, Following a binary logistic distribution, P represents the probability of an event occurring, and 1-P represents the probability of it not occurring. Since the research data Y only contains the numbers 1 and 0, when building the model, the original data is 0, and the predicted value fitted by the model will be as close to 0 as possible. X1, X2, etc., on the right side of the equation are the research objects or influencing factors, specifically the Points of Interest (POIs); while γ1, γ2, etc., are influence coefficients, representing the degree of influence of each influencing factor on the research data.

[0031] By analyzing the spatial correlation between the number of clusters in each cluster and the number of the base cluster (the group with the most data) in each cluster data set using regression models, the correlation between location characteristics and the layout of shared bicycle parking can be obtained. For example, to determine whether changes in the number of residential areas, bus stops, restaurants, sports venues, and companies within a 300-meter radius of subway stations have a significant positive effect on the flow of shared bicycles entering the station.

[0032] The beneficial effects of this invention are as follows: This invention applies pattern recognition techniques such as cluster analysis to classify bicycle stations based on daily usage patterns, enriching domestic research methods on bicycle station classification and providing important references for understanding the classification and function of bicycle stations, thus providing effective basis for travelers to make better travel decisions. Through a binary logistic regression model, it identifies the relationship between station type and location characteristics while controlling for the spatial dependence of neighboring stations, filling a research gap in the interrelationships between station clusters. It provides a basis for better improving public transportation utilization and the traffic environment in city centers, and deepens the understanding of public bicycle travel characteristics, thereby enabling the rational allocation of public bicycle demand and improving the traffic environment. Furthermore, it utilizes a geographic information system to visualize the spatiotemporal dynamics of public bicycle use, providing clearer data support for domestic research on public bicycle travel. Attached Figure Description

[0033] Figure 1 A map showing the distribution of subway stations in Shanghai;

[0034] Figure 2 A map showing the distribution of bus stops in Shanghai;

[0035] Figure 3 A weekly map showing the distribution of shared bicycle usage in Shanghai;

[0036] Figure 4 This is a map showing the distribution of shared bicycle usage on August 27th.

[0037] Figure 5 This is a map showing the distribution of shared bicycle usage on September 1st.

[0038] Figure 6 A graph showing the relationship between the number of clusters and the average silhouette coefficient;

[0039] Figure 7 A spatial distribution map of shared bicycle stations around subway stations based on primary membership values;

[0040] Figure 8 A spatial distribution map of shared bicycle stations around bus stops based on primary membership values;

[0041] Figure 9A scatter plot showing the distribution of shared bicycles near bus stops (average: 176.64).

[0042] Figure 10 A scatter distribution map of shared bicycles near subway stations (average: 1357.14);

[0043] Figure 11 A scatter plot showing the arrival points of shared bicycles at a typical BSS (Bike Service Site) station (average: 74.24).

[0044] Figure 12 A scatter plot showing the distribution of shared bicycle departure points near bus stops (average value: 159.12).

[0045] Figure 13 A scatter plot of shared bicycle departure points near subway stations (average value: 1391.07);

[0046] Figure 14 This is a scatter plot showing the distribution of shared bicycle departure points at a typical BSS (Bike Support System) station (average value: 73.60). Detailed Implementation

[0047] This invention proposes a clustering-based method for determining the spatiotemporal layout of shared bicycles. Figure 1 and Figure 2 Taking the road network and parking lot layout diagram shown as an example, this illustrates the method for determining the spatiotemporal layout of shared bicycle activities.

[0048] 1. Data analysis and processing of shared bicycle travel.

[0049] The spatiotemporal distribution of public bicycles in Shanghai was selected as a case study in this research. Real-time data on Hello Bike usage in Shanghai for the week of August 27th to September 2nd, 2018, was obtained from relevant personnel at Hello Bike. Figure 3 As shown in the diagram. Based on the distribution maps of subway stations and bus stops, a traffic area with a radius of 300 meters is divided with the subway station and bus stop as the center point. Each traffic area is regarded as a "fixed station", and a fuzzy clustering model is established accordingly.

[0050] For each trip, four elements were selected for data collection: arrival (locking) / departure (unlocking) time, latitude and longitude location, lock status, and bicycle number. The data collection system collected a dataset of approximately 23 million data points. To properly utilize the data within the dataset, it was filtered, processed, and transformed to create a new dataset applicable to research and analysis.

[0051] The trends of shared bike arrivals and departures at stations reveal the main spatial changes throughout the day. According to the chart, the daily arrival and departure volumes of shared bikes in Shanghai are relatively balanced. However, the distribution of these volumes over a week is unbalanced, roughly divided into two categories: weekday data and weekend data. The chart shows that the flow and density of shared bike departures and arrivals in Shanghai are significantly higher on weekdays than on weekends.

[0052] Figure 4 and Figure 5 The chart shows the distribution of inbound / outbound traffic for each hour on a selected weekday (Monday, August 27, 2018) and weekend (Saturday, September 1, 2018). It reveals two peak periods for inbound / outbound traffic on weekdays: 7:00 AM to 10:00 AM and 5:00 PM to 8:00 PM. The number of inbound / outbound activities during these peak periods is significantly higher than during normal hours. On weekends, there is only one significant peak period for inbound / outbound traffic, from 8:00 AM to 10:30 AM, followed by a continuous decline over time. Compared to the weekday pattern, the weekend travel pattern shown in the chart exhibits less inbound / outbound traffic, and the demand distribution throughout the day is more spatially balanced and dispersed.

[0053] First, the analysis conditions were defined: 440 valid subway stations, 989 valid bus stations, 400 randomly selected ordinary BSS stations, morning peak hours from 7:00 to 9:00, and evening peak hours from 17:30 to 19:30. Second, before applying the fuzzy clustering algorithm, a buffer zone was established using ArcMap software (a study area with a radius of 300 meters centered on each station). The analysis time periods were divided into weekdays and weekends. Spatial cross-analysis was performed on the entry and exit events of shared bicycles and the buffer zones of the two types of stations to obtain all entry / exit data streams, morning peak entry / exit data streams, and evening peak entry / exit data streams. All data were divided into 8 groups: weekday exit data around subway stations, weekday entry data around subway stations, weekday exit data around bus stations, weekday entry data around bus stations, weekend exit data around subway stations, weekend entry data around subway stations, weekend exit data around bus stations, and weekend entry data around bus stations. Finally, the entry and exit data are integrated into three sets of overall analysis data. Taking weekday subway stations as an example, the data format is shown in Table 1.

[0054] Table 1 Spatial Cross-Analysis Summary Table

[0055]

[0056] Based on all the obtained data, the average traffic flow every ten minutes for shared bicycles entering and exiting all stations was calculated to facilitate analysis and clustering. In calculating the average traffic flow every ten minutes for all entering and exiting stations, the traffic flow from 00:30 to 04:30 was considered negligible, and 20 hours were selected as the effective time period within a day. Based on the average traffic flow every ten minutes for all entering and exiting stations, the ratio of the average traffic flow every ten minutes during the morning peak and evening peak was calculated, denoted as the morning peak entry / exit ratio and evening peak entry / exit ratio. The corresponding traffic flow ratios for the morning and evening peak periods were also calculated. For example, data from weekday subway stations is shown in Tables 2 and 3.

[0057] Table 2. Weekday Arrival Data Compilation Table for Subway Stations

[0058]

[0059]

[0060] Table 3. Weekday Departure Data from Subway Stations (Compiled Table)

[0061]

[0062] 2. Cluster analysis

[0063] Statistical analysis was performed on each station based on the average number of times shared bicycles entered / exited every ten minutes throughout the day, the entry / exit ratio during the morning peak, and the entry / exit ratio during the evening peak. A C-means clustering algorithm based on minimizing the weighted squared error function was then applied.

[0064] Tables 5-12 depict the inbound and outbound activity characteristics of various clusters obtained from fuzzy clustering analysis for each group. The classification properties of these stations are plotted based on the value of the primary cluster with the highest membership degree. Clusters can be distinguished based on the total daily flow and the temporal variation of the flow.

[0065] By writing code using Python software, a graph showing the relationship between the number of clusters and the average silhouette coefficient can be obtained, such as... Figure 6 As shown (taking weekday departure data from around subway stations as an example).

[0066] According to the relationship diagram, when the silhouette coefficient reaches its maximum value, the value of k equals 3, which indicates that the optimal number of clusters is 3. However, it is worth noting that, combining the data of k and SSE, we can see that when k is 3, the SSE is still very large, so this is not a very reasonable number of clusters. Considering the second largest silhouette coefficient, k is 5, and similarly, the SSE is already at a relatively low level at this time. Therefore, the optimal cluster coefficient should be 5.

[0067] After contour analysis and processing of all groups, the final results are shown in Table 4:

[0068] Table 4. Cluster Count for Each Group

[0069]

[0070] Table 5. Cluster Analysis of Shared Bike Departure Data Around Subway Stations on Weekdays

[0071]

[0072]

[0073] Table 6. Cluster Analysis of Shared Bike Arrival Data Around Subway Stations on Weekdays

[0074]

[0075] Table 7. Cluster Analysis of Shared Bike Departure Data Around Bus Stops on Weekdays

[0076]

[0077]

[0078] Table 8. Cluster Analysis of Shared Bike Departure Data Around Bus Stops on Weekdays

[0079]

[0080] Table 9. Cluster Analysis of Shared Bike Departure Data Around Subway Stations on Weekends

[0081]

[0082]

[0083] Table 10 Cluster Analysis of Shared Bike Arrival Data Around Subway Stations on Weekends

[0084]

[0085] Table 11 Cluster Analysis of Shared Bike Departure Data Around Bus Stops on Weekends

[0086]

[0087]

[0088] Table 12 Cluster Analysis of Shared Bike Arrival Data Around Bus Stops on Weekends

[0089]

[0090] The table above describes the main statistical information of various clusters within each category. It can be observed that on weekdays, exit data around subway stations shows peak periods in the morning and evening, with a general trend of morning peak exit bicycle traffic exceeding evening peak traffic. This indicates that subway stations exhibit spatial and temporal clustering for shared bicycle travel. Furthermore, by analyzing each cluster separately, we can identify their distinct characteristics:

[0091] The stations within Cluster 1 maintain high outbound traffic throughout the day, exhibiting significant peaks not only in the morning and evening, but also with morning peak outbound traffic far exceeding evening peak outbound traffic. These stations can be categorized as "high-demand stations with a focus on morning peak outbound traffic." Therefore, during the morning peak hours, the demand for shared bicycles at stations within Cluster 1 typically exceeds the demand for parking spaces.

[0092] Cluster 2 exhibits a similar distribution trend to Cluster 1 throughout the day, but its outbound traffic remains at a moderate level throughout the day. This type of site can be categorized as a "site with moderate demand, particularly during the morning peak outbound traffic period".

[0093] The stations in Cluster 3 show significant differences in peak hours between the morning and evening, with the number of passengers leaving the station during the morning peak being significantly higher than during the evening peak. However, the overall outbound traffic remains very low throughout the day. These stations can be labeled as "low-demand stations that focus on exiting during the morning peak".

[0094] The stations in Cluster 4 maintain high outbound traffic throughout the day, exhibiting significantly similar peaks in the morning and evening. These stations can be labeled as "double-peak outbound high-demand stations".

[0095] The stations in Cluster 5 only show significantly similar peaks in the morning and evening, but the outbound traffic remains very low throughout the day. These stations can be labeled as "dual-peak outbound low-demand stations." Working with station clusters helps to understand the usage patterns of individual stations in terms of quantity, variation, and the balance of inbound / outbound activities.

[0096] Based on the data in the table, the exit data around subway stations on weekdays shows a peak period in the morning and evening, with the morning peak showing a greater flow of bicycles exiting than the evening peak. This indicates that subway stations have both spatial and temporal clustering effects on shared bicycle travel. In the analysis of exit data around bus stations on weekdays, the clustering effect of bus stations is weaker than that of subway stations. Even without considering the fact that there is almost no exit or entry traffic throughout the day, the data still shows a peak period in the morning and evening, indicating a significant temporal clustering effect.

[0097] Figure 7This displays a spatial distribution map of bicycle-sharing stations around subway stations on weekdays and weekends, based on master membership values ​​for exit and entry data. Figure 8 This image shows a spatial distribution map of shared bicycle stations around bus stops on weekdays and weekends, based on their departure and arrival data and master membership values. The map reveals different clustering patterns with varying locational characteristics, warranting further investigation. Through comparison and analysis... Figure 7 and Figure 8 Based on the spatial distribution map of bicycle sharing stations using master membership values, the following conclusions can be drawn:

[0098] (1) Each station has its own characteristics in terms of the number of clusters and the distribution of cluster points based on arrival data and departure data in different time periods.

[0099] (2) Through Figure 7 It can be observed that although the cluster distribution of shared bicycle stations around subway stations varies at different times, the overall distribution pattern remains basically consistent; through Figure 8 It can be found that when selecting effective stations within the bus stop area, the distribution and number of effective stations for entering / exiting the station on weekdays and weekends are different, which is due to the different travel time periods. Through horizontal comparative research, compared with the cluster distribution of shared bicycle stations around subway stations, the cluster distribution of shared bicycle stations around bus stations is more irregular and more dispersed.

[0100] Figures 9 to 11 The demonstration showcased shared bicycles on public transportation ( Figure 9 ),subway( Figure 10 ) and regular BSS sites ( Figure 11 The arrival distribution state, Figures 12 to 14 The demonstration showcased shared bicycles on public transportation ( Figure 12 ),subway( Figure 13 ) and regular BSS sites ( Figure 14 The distribution of shared bicycles was analyzed. By comparing the distribution patterns and average data, and by comparing with ordinary BSS stations, it can be clearly found that both subway stations and bus stations have a time and space aggregation effect on the number of shared bicycle trips. Moreover, subway stations have a higher intensity and wider range of aggregation effect on shared bicycle trips.

[0101] 3. Regression Analysis

[0102] By organizing relevant geographic information data within a 300-meter radius of subway stations and bus stations obtained from the data cloud platform of Gaode Map Open Platform, and based on the data results obtained from the previous fuzzy clustering, the number of clusters with the largest number in each group of cluster data is selected as the basic cluster number for analysis, and set to Y=0.

[0103] We set the number of other cluster types to Y=1 and used Python to build a binary logistic regression model to analyze the spatial correlation between the number of other cluster types and the basic cluster type. The data in Table 13 (taking weekday entry data around subway stations as an example) was obtained.

[0104] Table 13 Regression Analysis of Shared Bicycle Arrival Data Around Subway Stations on Weekdays

[0105]

[0106]

[0107] The data in the table above mainly includes estimated values ​​of the coefficients and the significance level of the data fit. The following formula is used for the significance level analysis:

[0108] {P>|z|}=1-α

[0109] Where P represents the probability of an event occurring, z represents the critical limit for judging the truth or falsehood of the event, P>|z| represents the probability of accepting the null hypothesis when it is true in hypothesis testing, α is the significance level, and 1-α is called the confidence level.

[0110] Factors with α < 0.1 were selected as significant variables and included in the model; insignificant factors with α ≥ 0.1 were excluded from the model. This yields the specific form of the regression analysis model for weekday subway station entry data based on cluster 1 as the base cluster.

[0111] The specific form of the binary logistic regression model of cluster 2 on cluster 1 is as follows:

[0112]

[0113] P0 represents the probability of the arrival / departure event of a shared bicycle in the basic cluster, and P1 represents the probability of the arrival / departure event of a shared bicycle in other clusters.

[0114] Analysis of the binary logistic regression model of cluster 2 against cluster 1 reveals a significant positive effect on the number of shared bicycles entering subway stations on weekdays, as well as changes in the number of residential areas, bus stops, restaurants, sports venues, and businesses within a 300-meter radius. In other words, an increase in the values ​​of these factors has a significant positive effect on the flow of shared bicycles entering subway stations on weekdays. The event has a positive effect. Regarding the number of individual residential areas, assuming other variables remain constant, for every additional residential area within a 300-meter radius, the logarithm of the area moving closer to cluster 2 from cluster 1 increases. Increased by 0.3012. Conversely, changes in the number of shopping services within a 300-meter radius have a significant negative effect on the flow of shared bicycles entering subway stations on weekdays.

[0115] Combining the cluster characteristics of clusters 1 and 2, we can see that an increase in the number of residential areas, bus stops, catering services, sports venues, and companies within a 300-meter radius will help increase the flow of shared bicycles entering subway stations on weekdays; however, an increase in the number of shopping services within a 300-meter radius will hinder the increase in the flow of shared bicycles entering subway stations on weekdays.

[0116] The specific form of the binary logistic regression model of cluster 3 on cluster 1 is as follows:

[0117]

[0118] The specific form of the binary logistic regression model of cluster 4 on cluster 1 is as follows:

[0119]

[0120] Tables 14 and 15 describe the regression analysis data obtained by combining exit / entry clustering results on weekdays and weekends, and around subway stations and bus stops.

[0121] Table 14: Data Regression Analysis Summary of Shared Bicycles Around Subway Stations

[0122]

[0123]

[0124] Table 15: Data Regression Analysis Summary of Shared Bicycles Around Bus Stops

[0125]

[0126]

[0127] The study of Table 14 revealed that the main geographical influencing factors on travel data around subway stations include schools, residential areas, shopping services, bus stations, catering services, sports venues, hotels, and companies. Secondly, the number of residential areas was the most significant factor in the entire analyzed geographic information system, while the number of companies also had a relatively significant impact on travel data around subway stations. Compared to weekday departures and arrivals, weekend departures and arrivals were influenced by a wider range of factors. Weekend departures were specifically linked to the number of Class A hospitals, indicating that travelers tended to return home after visiting a hospital on weekends; similarly, weekend arrivals were specifically linked to entertainment venues, suggesting that travelers preferred to spend time at entertainment venues on weekends.

[0128] The study of Table 15 reveals the following: First, the main geographical influencing factors for travel data around bus stops include schools, residential areas, shopping services, subway stations, sports venues, and companies. Second, the number of residential areas is the most significant factor in the entire analyzed geographic information system, while the number of companies also has a relatively significant impact on travel data around subway stations. Finally, there are specific connections between weekday departures and arrivals, and weekend departures and arrivals for travel data around bus stops. Weekday departures are specifically linked to schools, indicating a preference for commuting to and from school on weekdays; weekend departures are specifically linked to the number of top-tier hospitals, indicating a preference for returning home after hospital visits on weekends; and weekend arrivals are specifically linked to catering services, indicating a preference for dining out on weekends.

[0129] For groups with 5 clusters, three or more factors showing a significant correlation between the number of clusters and the basic cluster number were selected as significant influencing factors; for groups with 4 clusters, two or more factors showing a significant correlation between the number of clusters and the basic cluster number were selected as significant influencing factors. The analysis yielded Tables 16 and 17.

[0130] Table 16 Distribution of Influencing Factors in Regression Analysis of Data Around Subway Stations

[0131]

[0132]

[0133] Table 17 Distribution of Influencing Factors in Regression Analysis of Data Around Bus Stops

[0134]

[0135] Tables 16 and 17 describe the distribution of the influence factors of regression analysis based on clustering results for exiting and entering subway and bus stations during weekday and weekend hours.

[0136] According to the table information, for subway stations and bus stops, the number of residential areas consistently has the most significant impact on shared bicycle travel, while the number of companies and enterprises is a secondary significant factor. Each analysis group has its own specific influencing factors. For data around subway stations, weekday exit data shows a unique and significant correlation with the number of hotels and sports venues; weekday entry data shows a unique and significant correlation with the number of catering services and sports venues; weekend exit data shows a unique and significant correlation with the number of Grade A hospitals; and weekend entry data shows a unique and significant correlation with the number of sports venues. For data around bus stops, weekday entry data shows a unique and significant correlation with the number of subway stations and sports venues; and weekend exit data shows a unique and significant correlation with the number of subway stations.

[0137] In summary, while the daily arrival and departure volumes of shared bicycles in Shanghai are relatively balanced, their distribution over a week is unbalanced. Furthermore, the flow and density of shared bicycle arrivals and departures are significantly higher on weekdays than on weekends. A comparison with regular BSS (Bus Service Station) sites reveals that both subway and bus stops have a significant temporal and spatial aggregation effect on shared bicycle usage, with subway stations exhibiting a more intense and widespread aggregation effect. Regardless of whether it's a subway or bus stop, weekday or weekend, or whether entering or exiting the station, the number of residential areas is the most significant influencing factor on shared bicycle usage, while the number of companies and enterprises is a secondary significant factor. Other influencing factors exhibit characteristics varying with time periods and geographical locations.

Claims

1. A clustering-based method for determining the spatiotemporal layout of shared bicycles, characterized in that... Includes the following steps: S1. Data analysis and processing of shared bicycle travel; Using the time and latitude / longitude of shared bicycles entering and leaving the station as basic data, a transportation area is established with subway and bus stations as the center. Using ArcMap software, a buffer zone was established, that is, a research area with a radius of h meters was established with each station as the center. The analysis time period was divided into two categories: weekdays and weekends. Spatial cross-analysis was performed on the shared bicycle flow entering and leaving the station and the station buffer zone range to obtain all data streams of entering or leaving the station, as well as the data streams of entering or leaving the station during the morning peak and the data streams of entering or leaving the station during the evening peak. The data on all shared bicycles entering and leaving the public bicycle sharing system (BSS) stations are divided into 8 groups, and then the data on entering or leaving the station are integrated into three groups of overall analysis data. Based on all the data obtained, calculate the average traffic per gram per minute for the shared bicycle entry and exit data of all BSS sites; Based on the average flow rate per g minute for all BSS inbound or outbound passengers, calculate the ratio of the average flow rate per g minute for inbound or outbound passengers during the morning and evening peak hours to this ratio, and record them as the inbound or outbound morning peak ratio and evening peak ratio. Also calculate the ratio of the corresponding flow rates during the morning and evening peak hours. S2. Use fuzzy C-means clustering analysis to obtain the activity characteristics of each cluster; The fuzzy C-means clustering algorithm is used to cluster shared bicycle stations based on differences in daily usage time; During the clustering process, the time period was divided into weekdays and weekends, and the shared bicycle flow entering and leaving the BSS station was considered separately. After determining the predetermined number of clusters, three indicators were selected as the dissimilarity measurement objects: the average flow per g minutes of all entering or leaving the station, the ratio of morning peak to evening peak of entering or leaving the station, and the squared error was used as the dissimilarity measurement value. Modeling analysis was performed on the 8 groups. The silhouette coefficient method was selected as the main basis for determining the number of clusters in each group, and the sum of squared errors was used to determine the number of clusters in each group of data. Statistical analysis was performed on each site based on the average number of times shared bicycles entered / exited every g minutes throughout the day, the entry / exit ratio during the morning peak, and the entry / exit ratio during the evening peak. The C-means clustering algorithm based on minimizing the weighted square error function was applied to divide the shared bicycles into clusters based on the daily flow and time. The sites were then classified according to the main cluster value with the highest membership degree, and the activity characteristics of various shared bicycle clusters were obtained. S3. Regression analysis of shared bicycle location characteristics based on POI; We acquired relevant geographic information data within a h-meter radius of subway stations and bus stations, respectively. Through the data cloud platform of Gaode Map Open Platform, and by using Python software to build a data acquisition model, we obtained Points of Interest (POIs) within a h-meter radius of each station. A binary Logit model was used for regression analysis to obtain the locational characteristics of shared bicycle parking layout; Based on the continuous membership values ​​generated by the C-means clustering method, a binary logistic regression model can be used to explore the determinants of cluster membership while controlling for spatial autocorrelation between sites. The spatial correlation between the number of clusters in each cluster and the number of basic clusters in each cluster data was analyzed by using a binary logistic regression model to obtain the correlation between location characteristics and the layout of shared bicycle parking. In step S2, the fuzzy C-means clustering algorithm continuously optimizes the objective function to obtain the membership degree of each sample point to each cluster center, and selects the highest membership degree as the classification criterion, thereby achieving the purpose of automatically classifying the sample data. This clustering algorithm aims to solve the following objective function: Where c is the predetermined number of clusters, m is the fuzziness, which is a parameter that determines the influence of outliers on the centroid values ​​of clusters; and N is the number of discrete clusters. This represents the membership value (or degree) of the cluster from the i-th site to the k-th site. For a given site, the sum of its membership values ​​with all other clusters is 1. i , x k L(x) represents the value of the i-th object and the value of the k-th cluster centroid, respectively. i , x k () is a measure of the dissimilarity between the i-th object and the k-th cluster centroid value; Using the sum of squared residuals as a measure of anisotropy, the membership values ​​satisfy the following condition: At this point, the optimal conditions for clustering have been met.

2. The method for determining the spatiotemporal layout of shared bicycles based on clustering according to claim 1, characterized in that: In step S1, before data processing, the analysis conditions are determined: the number of valid subway stations, the number of valid bus stations, the number of randomly selected ordinary stations, and the morning and evening peak hours. In calculating the average flow of shared bicycles entering or leaving the station every ten minutes, the flow from 0:30 to 4:30 is ignored, and 20 hours are selected as the effective time period within a day.

3. The method for determining the spatiotemporal layout of shared bicycles based on clustering according to claim 1, characterized in that: The eight groups mentioned in step S1 are: weekday exit data around subway stations, weekday entry data around subway stations, weekday exit data around bus stops, weekday entry data around bus stops, weekend exit data around subway stations, weekend entry data around subway stations, weekend exit data around bus stops, and weekend entry data around bus stops.

4. The method for determining the spatiotemporal layout of shared bicycles based on clustering according to claim 1, characterized in that: In step S2, the core indicator of the profile coefficient method is the profile coefficient, which is the coefficient of a sample point. The contour coefficient is defined as follows: Where a is X i The average distance from other samples in the same cluster is called cohesion, where b is the distance of X. i The average distance to all samples in the nearest cluster is called the separation degree; The definition of the nearest cluster is as follows: Where p is a certain cluster C k The sample in the sample is X i After using the average distance to all samples in a cluster as a measure of the distance from a point to that cluster, select the distance from X. i The nearest cluster is designated as the nearest cluster; The average silhouette coefficient is obtained by calculating the silhouette coefficient of all samples according to the formula and then averaging them. Its value range is [-1, 1]. The k with the largest average silhouette coefficient is the optimal number of clusters.

5. The method for determining the spatiotemporal layout of shared bicycles based on clustering according to claim 1, characterized in that: In step S3, the formula for the binary logistic regression model is established as follows: In the above formula, It follows a binary logistic distribution, where P represents the probability of the event occurring, 1-P represents the probability of the event not occurring, and X1, X2… on the right side of the equation are… All of these are research subjects or influencing factors, specifically points of interest (POIs). , … All are influence coefficients, representing the degree of influence of each influencing factor on the research data.

6. The method for determining the spatiotemporal layout of shared bicycles based on clustering according to claim 5, characterized in that: The POIs include schools, residential areas, food and beverage services, sports venues, shopping services, and businesses.

Images