Rail transit connection shared bicycle use influence factor analysis method based on built environment and interpretable machine learning

Abstract

The invention relates to the technical field of urban and rural traffic planning, in particular to a rail transit connection shared bicycle use influence analysis method based on built environment and interpretable machine learning, which comprises the following steps: S1, identifying traffic behaviors of taking shared bicycles as rail transit connection trips, obtaining shared bicycle order data in a target area, and determining the order data of the shared bicycles in the target area; constructing a shared bicycle connection use database; s2, determining a research area of a built environment around each subway station, and delimiting an annular buffer area around the subway station as a research range; s3, fusing multi-source built-up environment data, including objective built-up environment indexes and perceptual built-up environment indexes; s4, on the basis of the data in S1 and S3, establishing an influence factor data set of the environment on the use of the shared bicycle; the method comprises the following steps: identifying a traffic behavior of taking a shared bicycle as a rail traffic connection trip, obtaining shared bicycle order data in a target area, extracting riding OD point, distance and time information, and constructing a shared bicycle connection use database;

Description

Technical Field

[0001] The present invention relates to the technical field of urban and rural transportation planning, and specifically to a method for analyzing the impact of rail transit connection shared bicycle usage based on built environment and explainable machine learning. Background Art

[0002] With the acceleration of urbanization, the growth of urban populations has placed higher demands on urban public transportation systems. Urban rail transit, as a key component of urban transportation networks, has played a vital role in improving urban transportation efficiency. However, its limited station coverage and network density have led to the widespread "last mile" problem. This challenge not only affects residents' travel experience but also limits the attractiveness and efficiency of rail transit.

[0003] To address this challenge, the Bike-and-Ride model, which combines rail transit and bicycles, is considered an effective strategy to increase rail transit's travel share and a key model for solving the "last mile" problem of urban rail transit. Bike sharing (BS), with its flexibility and convenience, has become an indispensable part of urban public transportation systems. Bike sharing not only effectively solves the "last mile" travel problem, but also promotes the coordinated development of rail transit and green travel.

[0004] However, shared bicycles also face many challenges in the process of connecting with rail transit, such as uneven temporal and spatial distribution, lack of non-motorized vehicle-only roads and parking facilities, improper land use and public transportation facility layout, and other unreasonable built environment factors, which may affect the continuity, comfort and connection efficiency of riding. This is not only not conducive to the healthy development of shared bicycles, but also hinders the full realization of the benefits of rail transit. Therefore, a deep understanding of the factors affecting cycling in the built environment around rail transit stations, and then planning and designing a good cycling environment around stations, to increase residents' cycling willingness and the travel share of urban public transportation, has become a hot topic in transportation research in recent years. Exploring the impact mechanism of the built environment around rail transit stations on shared bicycle travel through massive shared bicycle riding data is of great significance for the optimization and improvement of current urban rail transit connection facilities.

[0005] After searching, the Chinese patent database disclosed patents such as application number 202311146139.2, invention title: Shared bicycle usage impact assessment method integrating spatial heterogeneity and nonlinearity, and application number 202310051935.1, invention title: Analysis method of influencing factors of subway station and dockless shared bicycle connection travel;

[0006] However, the above invention still has the following limitations:

[0007] 1. Ambiguous research scope: Traditional methods rely on a fixed-radius buffer zone (e.g., 800 meters) to define the analysis scope, failing to consider differences in urban density and the cumulative distribution of cycling behavior, resulting in unscientific definition of built environment research space.

[0008] 2. Single data dimension: Existing studies mostly rely on data such as point of interest (POIs) and lack street view imagery, making it difficult to quantify the impact of spatial quality (such as green visibility and street interface continuity) on cycling behavior.

[0009] 3. Insufficient model explanation: Existing studies often use linear models such as linear regression and stepwise regression, which do not adequately consider nonlinear effects. This can lead to underestimation or overestimation of the impact of certain indicators when quantifying the synergistic effects between variables. The assumption of a pre-set linear relationship can lead to miscalculations of the impact of explanatory variables within certain ranges, resulting in a lack of actionability in strategic recommendations.

[0010] In view of this, a method for analyzing the factors affecting the use of rail transit-connected shared bicycles based on the built environment and explainable machine learning is provided to overcome the above-mentioned shortcomings. Summary of the Invention

[0011] The purpose of this invention is to address the shortcomings of the existing technology and propose a method for analyzing the factors affecting the use of rail transit shuttle shared bicycles based on the built environment and interpretable machine learning;

[0012] To achieve the above objectives, the present invention adopts the following technical solution: a method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on the built environment and interpretable machine learning, comprising:

[0013] S1: Identify the traffic behavior of using shared bicycles as rail transit connections, obtain the shared bicycle order data in the target area, and build a shared bicycle connection usage database;

[0014] S2: Determine the study area of ​​the built environment around each subway station and define a circular buffer zone around the subway station as the study scope;

[0015] S3: Integrate multi-source built environment data, including objective built environment indicators and perceived built environment indicators;

[0016] S4: A dataset of factors influencing the built environment on the use of shared bicycles is constructed based on the data of S1 and S3;

[0017] S5: Use XGBoost model to train and determine model performance;

[0018] S6: The SHAP explanatory model is used to analyze the importance ranking, nonlinear effects and interaction mechanisms of built environment factors.

[0019] Furthermore, in the present invention, the specific method for identifying connecting travel behavior in step S1 includes: creating a buffer zone at each entrance and exit of the subway station, screening and analyzing the parking data of shared bicycles, reconstructing the travel chain, and using the KD-Tree spatial search algorithm to identify connecting travel behavior.

[0020] Furthermore, in the present invention, the method for defining the research scope in step S2 includes: performing kernel density estimation on the riding distance of rail transit connecting shared bicycles, and using the walking and cycling connection thresholds and the 85th percentile value to define the cycling connection research scope.

[0021] Furthermore, in the present invention, the objective built environment data in the multi-source data set in step S3 includes population density / employment density, land use combination / job-housing balance, POIs information, road network data, bus station data and subway station data; the perceived built environment data includes street greening rate, boundary enclosure, complexity and safety.

[0022] Furthermore, in the present invention, when the XGBoost model is used for training in step S5, the multi-source data set is randomly divided into a training set and a test set, a gradient boosting decision tree is used for model training, and the performance of the prediction model is evaluated by the mean square error, root mean square error, mean absolute error and determination coefficient.

[0023] Furthermore, in the present invention, when the SHAP explanatory model is used in step S6 to analyze the importance ranking, nonlinear effects and interaction mechanisms of built environment factors, the importance quantification value of each feature is calculated and a feature dependency graph is output.

[0024] Furthermore, in the present invention, the influencing factor dataset in step S4 specifically includes:

[0025] S4.1: Count the number of shared bicycles used. Use the latitude and longitude information of the shared bicycles at the start of use as the location information of shared bicycle use. Calculate the average number of shared bicycles used at each subway station during the study period as the number of shared bicycles used at each subway station. The calculation is shown in formula (1):

[0026]

[0027] Among them, Bike_use ID＝i is the average number of shared bicycles used in subway station i, is the number of shared bicycles used in subway station i on day n, where N is the total number of days in the study period;

[0028] S4.2: Count the population density within the buffer zone of each subway station and calculate the population density index of the built environment characteristics, as shown in formula (2):

[0029]

[0030] Among them, people_density ID＝i is the population density in subway station i, people i is the population of subway station i, area i is the buffer area of ​​subway station i;

[0031] S4.3: Count the employment population density within the buffer zone of each subway station and calculate the employment density index of the built environment characteristics, as shown in formula (3):

[0032]

[0033] Among them, employment_density ID＝i is the population density in subway station i, employment i is the population of subway station i, area i is the buffer area of ​​subway station i;

[0034] S4.3: Count the land use mix within the buffer zone of each subway station and use the entropy index model to calculate the land use mix index of the built environment walkability, as shown in formula (4):

[0035]

[0036] Among them, Land_use ID＝i is the land use entropy within subway station i, is the proportion of the area of ​​the jth land use type within metro station i to the total area, and Z is the total number of land types within metro station i;

[0037] S4.4: Count the job-housing balance within the buffer zone of each subway station and calculate the job-housing balance index of the built environment characteristics, as shown in formula (5):

[0038]

[0039] Among them, JH_balance ID＝i is the job-housing balance within subway station i, Employment_POI i is the number of employment type POIs within subway station i, Resident_POI i is the POI value of the number of residents in metro station i;

[0040] S4.5: Count the POI information within the buffer zone of each subway station, including the number of restaurants, entertainment facilities, transportation facilities, and commercial facilities;

[0041] S4.6: Collect road network data within the buffer zone of each subway station, including the length of main roads, secondary roads, branch roads, sidewalks, and bicycle lanes;

[0042] S4.7: Collect bus stop and subway station data within the buffer zone of each subway station, including the number of bus stops / subway stations and the distance to the nearest bus stop / subway station;

[0043] S4.8: Count the street greening degree within the buffer zone of each subway station and calculate the greening degree index of the perceived built environment, as shown in formula (6):

[0044] GVI ID＝i =VI itree +VI iterrain Formula (6)

[0045] Among them, GVI ID＝i is the greening degree of the residential and working streets within metro station i, VI itree The pixel ratio of street view trees around subway station i, VI iterrain The pixel ratio of bushes in the street scene around metro station i;

[0046] S4.8: Count the street enclosure within the buffer zone of each subway station and calculate the enclosure index of the perceived built environment, as shown in formula (7):

[0047]

[0048] Among them, EVI ID＝i is the degree of street enclosure between work and residence within metro station i, VI iduilding The pixel ratio of streetscape buildings around subway station i, VI iwall Pixel ratio of street view walls around subway station i, VI itree The pixel ratio of street view trees around subway station i, VI iroad Pixel ratio of street view roads around subway station i, VI isidewalk Pixel ratio of street view sidewalks around subway station i,VI iterrain The pixel ratio of bushes in the street scene around metro station i;

[0049] S4.9: Count the street complexity within the buffer zone of each subway station and calculate the complexity index of the perceived built environment, as shown in formula (8):

[0050]

[0051] Among them, CVI ID＝i is the complexity of the work-residence street within subway station i, VI iperson The pixel ratio of people in street scenes around subway station i, VI isignThe pixel ratio of street scene road signs around subway station i, VI ilight Pixel ratio of street lighting around subway station i, VI itree The pixel ratio of street view trees around subway station i, VI iroad Pixel ratio of street view roads around subway station i, VI building The pixel ratio of street scene buildings around metro station i;

[0052] S4.10: Count the street safety within the buffer zone of each subway station and calculate the safety index of the perceived built environment, as shown in formula (9):

[0053] SVI ID＝i =VI iperson +VI isign +VI ilight +VI ifence Formula (9)

[0054] Among them, SVI ID＝i is the safety level of the work-residence street within subway station i, VI iperson The pixel ratio of people in street scenes around subway station i, VI isign The pixel ratio of street scene road signs around subway station i, VI ilight Pixel ratio of street lighting around subway station i, VI ifence The pixel ratio of street view fences around subway station i.

[0055] Furthermore, in the present invention, the specific prediction steps of step S5 are:

[0056] S5.1: Randomly split the multi-source dataset into two parts for model testing: 70% training set and 30% test set;

[0057] S5.2: Use gradient boosting decision trees. Implement ensemble learning based on decision trees using gradient boosting. The gradient boosting strategy returns a more robust prediction model in the form of continuous iteration of multiple weak learners, steadily improving the accuracy of the model.

[0058] S5.3: A model comparison experiment was conducted using the XGBoost framework to implement a gradient boosted decision tree using an efficient parallel training algorithm. The objective function of the XGBoost algorithm is shown in Equation (10):

[0059]

[0060] in, is the loss function, Ω(f p ) is a regularization term, including L1 regularization term and L2 regularization term; constant is a constant term;

[0061] S5.4: Evaluate the performance of the prediction model using four indicators: mean square error, root mean square error, mean absolute error, and coefficient of determination R2.

[0062] Furthermore, in the present invention, the specific prediction steps of step S5 are:

[0063] S5.1: Randomly split the multi-source dataset into two parts for model testing: 70% training set and 30% test set;

[0064] S5.2: Use gradient boosting decision trees. Implement ensemble learning based on decision trees using gradient boosting. The gradient boosting strategy returns a more robust prediction model in the form of continuous iteration of multiple weak learners, steadily improving the accuracy of the model.

[0065] S5.3: A model comparison experiment was conducted using the XGBoost framework to implement a gradient boosted decision tree using an efficient parallel training algorithm. The objective function of the XGBoost algorithm is shown in Equation (10):

[0066]

[0067] in, is the loss function, Ω(f p ) is a regularization term, including L1 regularization term and L2 regularization term; constant is a constant term;

[0068] S5.4: Evaluate the performance of the prediction model using four indicators: mean square error, root mean square error, mean absolute error, and coefficient of determination R2.

[0069] Furthermore, in the present invention, the specific prediction steps of step S6 are:

[0070] S6.1: Use the SHAP method to attribute the changes in the gradient boosting decision tree model to each independent variable. In a set of independent variables, the difference in the prediction of this independent variable is the marginal contribution. The Shapley value can be used to estimate the marginal contribution effect of each feature. By counting the size of the Shapley value of each independent variable, the importance quantification value of the independent variable is obtained. The calculation method is shown in the following formula (11):

[0071]

[0072] Where, Φ p represents the contribution of feature p; P represents the number of features; f(x) is the output of the original model; Z' represents the number of non-zero items in Z'; refers to all Z' vectors in the subset of non-zero entries in x';

[0073] S6.2: Output the summary distribution of independent variables with high importance quantification values, and identify the relationship between the value of each independent variable and its impact on the prediction based on the Shapley value. At the same time, output the feature dependency graph, consider the interaction of multiple features, and deeply explore the intrinsic relationship between the built environment and shared bicycle use.

[0074] The present invention has the following beneficial effects:

[0075] In this invention, by identifying the traffic behavior of using shared bicycles as rail transit connections, the shared bicycle order data in the target area is obtained, the riding OD point, distance, and time information are extracted, and a shared bicycle connection usage database is constructed;

[0076] In this paper, the built environment study area around each subway station is determined. In order to more scientifically define the built environment study range, a heuristic idea is proposed to define the study range of the circular buffer zone around the subway station based on the 85th percentile of the cumulative distribution of cycling distance and the walking connection range.

[0077] In this paper, multi-source built environment data are integrated, including objective built environment indicators (density, diversity, road network design, public transportation facilities and destination accessibility) and perceived built environment indicators (street greening rate, enclosure, complexity and safety indicators) to ensure that the research indicators are comprehensive and effective;

[0078] In the present invention, a dataset of factors affecting the use of shared bicycles by the built environment is constructed based on the data from steps 1 and 3;

[0079] In the present invention, the XGboost model is used for training, fine-tuning is performed on the validation set, and testing is performed on the test set to determine the model performance;

[0080] In this paper, the SHAP explanatory model is used to analyze the importance ranking, nonlinear effects and interaction mechanisms of built environment factors, and to propose more specific urban planning and connection optimization strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 Schematic diagram of the overall structure of the system in the present invention;

[0082] Figure 2 This is a diagram illustrating the connection of the system of the present invention;

[0083] Figure 3 This is a cycling distance distribution diagram in the present invention;

[0084] Figure 4 A diagram showing the steps of street scene recognition in the present invention;

[0085] Figure 5 This is a schematic diagram of the importance ranking of built environment factors in machine learning in the present invention;

[0086] Figure 6 This is a nonlinear relationship diagram used by the machine learning of the main roads and transportation facility POIs and shared bicycle connections in the present invention;

[0087] Figure 7 This is a schematic diagram of the results of the partial built environment interaction graph of machine learning in the present invention. DETAILED DESCRIPTION

[0088] With the acceleration of urbanization, urban rail transit, as a key component of the urban transportation network, has played a vital role in improving urban transportation efficiency. However, the limitations of rail transit station coverage and line network density have caused the ubiquitous "last mile" problem. In order to solve this problem, the Bike-and-Ride model came into being, that is, a travel mode that combines rail transit and bicycles. Among them, shared bicycles, with their flexibility and convenience, have become the key to solving the "last mile" problem. However, shared bicycles also face many challenges in the process of connecting with rail transit, such as uneven temporal and spatial distribution, lack of non-motorized vehicle dedicated roads and parking facilities, and other unreasonable built environment factors, which may affect the continuity, comfort and connection efficiency of riding. Therefore, the present invention proposes a method for analyzing the influencing factors of rail transit connection shared bicycle use based on built environment and interpretable machine learning, aiming to deeply understand the influencing factors of the built environment around the current rail transit station on riding, and then plan and design a good riding environment around the station, to improve residents' riding willingness and the travel share rate of urban public transportation.

[0089] A method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning, the method comprising:

[0090] S1: Identify the traffic behavior of using shared bicycles as rail transit connections, obtain shared bicycle order data in the target area, extract the riding OD point, distance, and time information, and build a shared bicycle connection usage database;

[0091] The specific method of identification in step S1 is:

[0092] S1.1: Create a 100-meter buffer zone at each entrance and exit of subway stations to filter and analyze shared bike parking data;

[0093] S1.2: Reconstruct the travel chain of shared bicycle parking data and build a travel OD data table;

[0094] S1.3: Use the KD-Tree (k-dimensional Tree) spatial search algorithm to perform a nearest neighbor search for each origin and destination in the trip OD data table to find the nearest subway station, identify connecting trips, and calculate their spatial distance to the subway station.

[0095] S1.4: Append the matched subway station ID and distance information to the OD data table to form a complete connecting trip dataset;

[0096] S2: Determine the study area of ​​the built environment around each subway station. Based on the 85th percentile of the cumulative distribution of cycling distance and the walking connection range, a 500-1800 m circular buffer zone around the subway station is defined as the study area.

[0097] The research scope is defined in step S2 as follows:

[0098] S2.1: Perform kernel density estimation on the riding distance of rail transit-connected shared bicycles and generate a cumulative distribution function graph of the riding distance;

[0099] S2.2: Use walking and cycling connection thresholds to define the lower limit of the cycling connection study scope;

[0100] S2.3: The 85th percentile value is a key indicator for defining the upper limit of the study range for cycling connections;

[0101] S3: Integrate multi-source built environment data, including objective built environment indicators (POI mix, road network density) and perceived built environment indicators (street greening rate, street openness);

[0102] In step S3, the objective built environment data in the multi-source dataset includes population density / employment density, land use mix / job-housing balance, POIs information, road network data, bus station data, and subway station data; the perceived built environment data includes street greening rate, boundary enclosure, complexity, and safety.

[0103] The multi-source built environment data collection method described in step S3 specifically includes:

[0104] S3.1: Select objective built environment based on the "5Ds" built environment, namely density, diversity, road network design, distance to transit, and destination accessibility; also include subway station-specific attribute indicators;

[0105] S3.2: The objective built environment uses multi-source big data, including Word Pop data, OpenStreetMap data, and Baidu Map data;

[0106] S3.3: Extract perceived built environment data based on Baidu Street View data. Specifically, based on the road network data, set a sample point every 50 meters on each road, and obtain street view images in four directions (0°, 90°, 180°, and 270°) at each sample point.

[0107] S3.4: Perform semantic image segmentation using deep learning techniques to calculate the pixel percentage in each street view image. Then, average the pixel percentages in the street view images in the four directions of a given sample point to obtain the final pixel percentage. Also, calculate the street greening rate, enclosure, complexity, and safety indicators.

[0108] S4: Constructing a dataset of factors affecting the built environment on the use of shared bicycles based on the data from steps S1 and S3; the dataset of factors affecting the built environment in step S4 specifically includes:

[0109] S4.1: Count the number of shared bicycles used. Use the latitude and longitude information of the shared bicycles at the start of use as the location information of shared bicycle use. Calculate the average number of shared bicycles used at each subway station during the study period as the number of shared bicycles used at each subway station. The calculation is shown in formula (1):

[0110]

[0111] Among them, Bike_use ID＝i is the average number of shared bicycles used in subway station i, is the number of shared bicycles used in subway station i on day n, where N is the total number of days in the study period;

[0112] S4.2: Count the population density within the buffer zone of each subway station and calculate the population density index of the built environment characteristics, as shown in formula (2):

[0113]

[0114] Among them, PEOPLE_DENSI ID＝i People-density ID＝i people-density ID＝i is the population density in subway station i, people i is the population of subway station i, area i is the buffer area of ​​subway station i;

[0115] S4.3: Count the employment population density within the buffer zone of each subway station and calculate the employment density index of the built environment characteristics, as shown in formula (3):

[0116]

[0117] Among them, employment_density ID＝i is the population density in subway station i, employment i is the population of subway station i, area i is the buffer area of ​​subway station i;

[0118] S4.3: Count the land use mix within the buffer zone of each subway station and use the entropy index model to calculate the land use mix index of the built environment walkability, as shown in formula (4):

[0119]

[0120] Among them, Land_use ID＝i is the land use entropy within subway station i, is the proportion of the area of ​​the jth land use type within metro station i to the total area, and Z is the total number of land types within metro station i;

[0121] S4.4: Count the job-housing balance within the buffer zone of each subway station and calculate the job-housing balance index of the built environment characteristics, as shown in formula (5):

[0122]

[0123] Among them, JH_balance ID＝i is the job-housing balance within subway station i, Employment_POI i is the number of employment type POIs within subway station i, Resident_POI i is the POI value of the number of residents in subway station i;

[0124] S4.5: Count the POI information within the buffer zone of each subway station, including the number of restaurants, entertainment facilities, transportation facilities, and commercial facilities;

[0125] S4.6: Collect road network data within the buffer zone of each subway station, including the length of main roads, secondary roads, branch roads, sidewalks, and bicycle lanes;

[0126] S4.7: Collect bus stop and subway station data within the buffer zone of each subway station, including the number of bus stops / subway stations and the distance to the nearest bus stop / subway station;

[0127] S4.8: Count the street greening degree within the buffer zone of each subway station and calculate the greening degree index of the perceived built environment, as shown in formula (6):

[0128] GVI ID＝i =VI itree +VI iterrain Formula (6)

[0129] Among them, GVI ID＝i is the greening degree of the residential and working streets within metro station i, VI itree The pixel ratio of street view trees around subway station i, VI iterrain The pixel ratio of bushes in the street scene around metro station i;

[0130] S4.8: Count the street enclosure within the buffer zone of each subway station and calculate the enclosure index of the perceived built environment, as shown in formula (7):

[0131]

[0132] Among them, EVI ID＝i is the degree of street enclosure between work and residence within metro station i, VI iduilding The pixel ratio of streetscape buildings around subway station i, VI iwall Pixel ratio of street view walls around subway station i, VI itree The pixel ratio of street view trees around subway station i, VI iroad Pixel ratio of street view roads around subway station i, VI isidewalk Pixel ratio of street view sidewalks around subway station i,VI iterrain The pixel ratio of bushes in the street scene around metro station i;

[0133] S4.9: Count the street complexity within the buffer zone of each subway station and calculate the complexity index of the perceived built environment, as shown in formula (8):

[0134]

[0135] Among them, CVI ID＝i is the complexity of the work-residence street within subway station i, VI iperson The pixel ratio of people in street scenes around subway station i, VI isign The pixel ratio of street scene road signs around subway station i, VI ilight Pixel ratio of street lighting around subway station i, VI itree The pixel ratio of street view trees around subway station i, VI iroad Pixel ratio of street view roads around subway station i, VI building The pixel ratio of street scene buildings around metro station i;

[0136] S4.10: Count the street safety within the buffer zone of each subway station and calculate the safety index of the perceived built environment, as shown in formula (9):

[0137] SVI ID＝i =VI iperson +VI isign +VI ilight +VI ifence Formula (9)

[0138] Among them, SVI ID＝i is the safety level of the work-residence street within subway station i, VI iperson The pixel ratio of people in street scenes around subway station i, VI isign The pixel ratio of street scene road signs around subway station i, VI ilight Pixel ratio of street lighting around subway station i, VI ifence The pixel ratio of street view fences around subway station i;

[0139] S5: Train the XGBoost model, fine-tune it on the validation set, and test it on the test set to determine the model performance.

[0140] The specific prediction steps of step S5 are:

[0141] S5.1: Randomly split the multi-source dataset into two parts for model testing: 70% training set and 30% test set;

[0142] S5.2: Use gradient boosting decision trees. Implement ensemble learning based on decision trees using gradient boosting. The gradient boosting strategy returns a more robust prediction model in the form of continuous iteration of multiple weak learners, steadily improving the accuracy of the model.

[0143] S5.3: A model comparison experiment was conducted using the XGBoost framework to implement a gradient boosted decision tree using an efficient parallel training algorithm. The objective function of the XGBoost algorithm is shown in Equation (10):

[0144]

[0145] in, is the loss function, Ω(f p ) is a regularization term, including L1 regularization term and L2 regularization term; constant is a constant term;

[0146] S5.4: The performance of the prediction model is evaluated by four indicators: mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination R2.

[0147] S6: Use the SHAP explanatory model to analyze the importance ranking, nonlinear effects and interaction mechanisms of built environment factors and generate connection optimization strategies;

[0148] The specific prediction steps of step S6 are:

[0149] S6.1: Use the SHAP method to attribute the changes in the gradient boosting decision tree model to each independent variable. In a set of independent variables, the difference in the prediction of this independent variable is the marginal contribution. The Shapley value can be used to estimate the marginal contribution effect of each feature. By counting the size of the Shapley value of each independent variable, the importance quantification value of the independent variable is obtained. The calculation method is shown in the following formula (11):

[0150]

[0151] Where, Φ p represents the contribution of feature p; P represents the number of features; f(x) is the output of the original model; Z' represents the number of non-zero items in Z'; Pointing to x 'All Z' vectors in the subset of non-zero entries;

[0152] S6.2: Output the summary distribution of independent variables with high importance quantification values, and identify the relationship between the value of each independent variable and its impact on the prediction based on the Shapley value. At the same time, output the feature dependency graph, consider the interaction of multiple features, and deeply explore the intrinsic relationship between the built environment and shared bicycle use.

[0153] This paper proposes a method for analyzing factors influencing the use of shared bikes connected to rail transit stations based on the built environment and interpretable machine learning. This method comprehensively analyzes the impact of the built environment around rail transit stations on shared bike use by identifying shared bike docking behaviors, determining a research area, integrating multi-source built environment data, constructing a dataset of influencing factors, training and testing the model using XGBoost, and employing the SHAP interpretable model to analyze the importance ranking, nonlinear effects, and interaction mechanisms of built environment factors.

[0154] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning, characterized by: include: S1: Identify the traffic behavior of using shared bicycles as rail transit connections, obtain the shared bicycle order data in the target area, and build a shared bicycle connection usage database; S2: Determine the study area of ​​the built environment around each subway station and define a circular buffer zone around the subway station as the study scope; S3: Integrate multi-source built environment data, including objective built environment indicators and perceived built environment indicators; S4: A dataset of factors influencing the built environment on the use of shared bicycles is constructed based on the data of S1 and S3; S5: Use XGBoost model to train and determine model performance; S6: The SHAP explanatory model is used to analyze the importance ranking, nonlinear effects and interaction mechanisms of built environment factors.

2. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1 is characterized in that: The specific method for identifying connecting travel behaviors in step S1 includes: creating a buffer zone at each entrance and exit of the subway station, screening and analyzing the parking data of shared bicycles, reconstructing the travel chain, and using the KD-Tree spatial search algorithm to identify connecting travel behaviors.

3. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1 is characterized in that: The method for defining the research scope in step S2 includes: performing kernel density estimation on the riding distance of rail transit connecting shared bicycles, and using the walking and cycling connection thresholds and the 85th percentile value to define the cycling connection research scope.

4. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1 is characterized in that: The objective built environment data in the multi-source dataset in step S3 include population density / employment density, land use mix / job-housing balance, POIs information, road network data, bus station data, and subway station data; the perceived built environment data include street greening rate, boundary enclosure, complexity, and safety.

5. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1 is characterized in that: When using the XGBoost model for training in step S5, the multi-source data set is randomly divided into a training set and a test set, the model is trained using a gradient boosting decision tree, and the performance of the prediction model is evaluated by the mean square error, root mean square error, mean absolute error, and determination coefficient.

6. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1 is characterized in that: In step S6, the SHAP explanatory model is used to analyze the importance ranking, nonlinear effects, and interaction mechanisms of built environment factors, including calculating the importance quantification value of each feature and outputting a feature dependency graph.

7. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1, characterized in that: The influencing factor dataset described in step S4 specifically includes: S4.1: Count the number of shared bicycles used. Use the latitude and longitude information of the shared bicycles at the start of use as the location information of shared bicycle use. Calculate the average number of shared bicycles used at each subway station during the study period as the number of shared bicycles used at each subway station. The calculation is shown in formula (1): Among them, Bike_use ID＝i is the average number of shared bicycles used in subway station i, is the number of shared bicycles used in subway station i on day n, where N is the total number of days in the study period; S4.2: Count the population density within the buffer zone of each subway station and calculate the population density index of the built environment characteristics, as shown in formula (2): Among them, people_density ID＝i is the population density in subway station i, people i is the population of subway station i, area i is the buffer area of ​​subway station i; S4.3: Count the employment population density within the buffer zone of each subway station and calculate the employment density index of the built environment characteristics, as shown in formula (3): Among them, employment_density ID＝i is the population density in subway station i, employment i is the population of subway station i, area i is the buffer area of ​​subway station i; S4.3: Count the land use mix within the buffer zone of each subway station and use the entropy index model to calculate the land use mix index of the built environment walkability, as shown in formula (4): Among them, Land_use ID＝i is the land use entropy within subway station i, is the proportion of the area of ​​the jth land use type within metro station i to the total area, and Z is the total number of land types within metro station i; S4.4: Count the job-housing balance within the buffer zone of each subway station and calculate the job-housing balance index of the built environment characteristics, as shown in formula (5): Among them, JH_balance ID＝i is the job-housing balance within subway station i, Employment_POI i is the number of employment type POIs within subway station i, Resident_POI i is the POI value of the number of residents in subway station i; S4.5: Count the POI information within the buffer zone of each subway station, including the number of restaurants, entertainment facilities, transportation facilities, and commercial facilities; S4.6: Collect road network data within the buffer zone of each subway station, including the length of main roads, secondary roads, branch roads, sidewalks, and bicycle lanes; S4.7: Collect bus stop and subway station data within the buffer zone of each subway station, including the number of bus stops / subway stations and the distance to the nearest bus stop / subway station; S4.8: Count the street greening degree within the buffer zone of each subway station and calculate the greening degree index of the perceived built environment, as shown in formula (6): GVI ID＝i = VI itree + VI iterrain Equation (6) Among them, GVI ID＝i is the greening degree of the residential and working streets within metro station i, VI itree The pixel ratio of street view trees around subway station i, VI iterrain The pixel ratio of bushes in the street scene around metro station i; S4.8: Count the street enclosure within the buffer zone of each subway station and calculate the enclosure index of the perceived built environment, as shown in formula (7): Among them, EVI ID＝i is the degree of street enclosure between work and residence within metro station i, VI iduilding The pixel ratio of streetscape buildings around subway station i, VI iwall Pixel ratio of street view walls around subway station i, VI itree The pixel ratio of street view trees around subway station i, VI iroad Pixel ratio of street view roads around subway station i, VI isidewalk Pixel ratio of street view sidewalks around subway station i,VI iterrain The pixel ratio of bushes in the street scene around metro station i; S4.9: Count the street complexity within the buffer zone of each subway station and calculate the complexity index of the perceived built environment, as shown in formula (8): Among them, CVI ID＝i is the complexity of the work-residence street within subway station i, VI iperson The pixel ratio of people in street scenes around subway station i, VI isign The pixel ratio of street scene road signs around subway station i, VI ilight Pixel ratio of street lighting around subway station i, VI itree The pixel ratio of street view trees around subway station i, VI iroad Pixel ratio of street view roads around subway station i, VI building The pixel ratio of street scene buildings around metro station i; S4.10: Count the street safety within the buffer zone of each subway station and calculate the safety index of the perceived built environment, as shown in formula (9): SVI ID＝i = VI iperson + VI isign + VI ilight + VI ifence Equation (9) Among them, SVI ID＝i is the safety level of the work-residence street within subway station i, VI iperson The pixel ratio of people in street scenes around subway station i, VI isign The pixel ratio of street scene road signs around subway station i, VI ilight Pixel ratio of street lighting around subway station i, VI ifence The pixel ratio of street view fences around subway station i.

8. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1, characterized in that: The specific prediction steps of step S5 are: S5.1: Randomly split the multi-source dataset into two parts for model testing: 70% training set and 30% test set; S5.2: Use gradient boosting decision trees. Implement ensemble learning based on decision trees using gradient boosting. The gradient boosting strategy returns a more robust prediction model in the form of continuous iteration of multiple weak learners, steadily improving the accuracy of the model. S5.3: A model comparison experiment was conducted using the XGBoost framework to implement a gradient boosted decision tree using an efficient parallel training algorithm. The objective function of the XGBoost algorithm is shown in Equation (10): in, is the loss function, Ω(f p ) is a regularization term, including L1 regularization term and L2 regularization term; constant is a constant term; S5.4: Evaluate the performance of the prediction model using four indicators: mean square error, root mean square error, mean absolute error, and coefficient of determination R2.

9. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1, characterized in that: The specific prediction steps of step S5 are: S5.1: Randomly split the multi-source dataset into two parts for model testing: 70% training set and 30% test set; S5.2: Use gradient boosting decision trees. Implement ensemble learning based on decision trees using gradient boosting. The gradient boosting strategy returns a more robust prediction model in the form of continuous iteration of multiple weak learners, steadily improving the accuracy of the model. S5.3: A model comparison experiment was conducted using the XGBoost framework to implement a gradient boosted decision tree using an efficient parallel training algorithm. The objective function of the XGBoost algorithm is shown in Equation (10): in, is the loss function, Ω(f p ) is a regularization term, including L1 regularization term and L2 regularization term; constant is a constant term; S5.4: Evaluate the performance of the prediction model using four indicators: mean square error, root mean square error, mean absolute error, and coefficient of determination R2.

10. The method for analyzing factors affecting the use of rail transit shuttle shared bicycles based on built environment and explainable machine learning according to claim 1, characterized in that: The specific prediction steps of step S6 are: S6.1: Use the SHAP method to attribute the changes in the gradient boosting decision tree model to each independent variable. In a set of independent variables, the difference in the prediction of this independent variable is the marginal contribution. The Shapley value can be used to estimate the marginal contribution effect of each feature. By counting the size of the Shapley value of each independent variable, the importance quantification value of the independent variable is obtained. The calculation method is shown in the following formula (11): Where, Φ p represents the contribution of feature p; P represents the number of features; f(x) is the output of the original model; Z' represents the number of non-zero items in Z'; refers to all Z' vectors in the subset of non-zero entries in x'; S6.2: Output the summary distribution of independent variables with high importance quantification values, and identify the relationship between the value of each independent variable and its impact on the prediction based on the Shapley value. At the same time, output the feature dependency graph, consider the interaction of multiple features, and deeply explore the intrinsic relationship between the built environment and shared bicycle use.

Images