A shared bicycle track connection analysis method considering built environment of travel stage

Abstract

The present application relates to the field of intelligent transportation system and big data analysis technology, and more particularly to a shared bicycle rail connection analysis method considering travel stage built environment, comprising: obtaining shared bicycle order trajectory data in a target area, identifying travel behavior, performing OD mapping of the identified connection travel starting point-destination, and generating shared bicycle travel OD samples; performing shared bicycle trajectory data corresponding to the OD samples for riding path reconstruction and path restoration; for different travel stages of shared bicycle connection travel, corresponding built environment features are extracted respectively; based on the built environment features of different travel stages, prediction models for predicting shared bicycle use probability of rail transit inbound connection and outbound connection are constructed respectively; the prediction models are explained and analyzed, and key influencing factors and their action relationships are output. The present application can more comprehensively and truly reflect the formation mechanism of shared bicycle and rail transit integrated travel.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation systems and big data analytics, and in particular to a method for analyzing shared bicycle rail connections that takes into account the built environment during the travel phase. Background Technology

[0002] With the continuous expansion of urban scale and the sustained growth of population, the rapid urbanization process has placed higher demands on urban public transportation systems. Urban rail transit, due to its large capacity, high efficiency, and good punctuality, has become an important mode of alleviating traffic congestion and supporting commuting. However, limited by station coverage and line density, rail transit generally faces the "first / last mile" connection problem, meaning that passengers experience inconvenience at the final stage of their journey, from their origin to the station or from the station to their destination, affecting the attractiveness and service efficiency of rail transit.

[0003] As a green, flexible, and economical mode of short-distance travel, shared bicycles provide an effective supplement to alleviating the aforementioned connection problems. Their "pick up anytime, return nearby" characteristic improves the accessibility and overall service level of rail transit. In recent years, the integration of shared bicycles and rail transit has become increasingly prominent. Related reports show that the average riding distance of shared bicycles around subway stations is about 1.4 kilometers, and the proportion of orders occurring near rail stations continues to rise, indicating that the "ride-sharing" model is playing an increasingly important role in increasing the proportion of public transportation use and solving last-mile transportation problems.

[0004] The built environment is a key factor influencing residents' travel behavior. Its characteristics, such as population density, employment density, land use mix, road network structure, and transportation infrastructure configuration, affect the choice of using shared bicycles for rail connections. Existing research has explored the mechanisms by which the built environment affects shared bicycle use and public transportation from multiple perspectives.

[0005] The existing technology still has the following shortcomings: 1. The analysis unit is too simple and does not distinguish between travel stages.

[0006] Existing studies on the integration of shared bicycles and rail transit mostly use rail transit stations as the unit of analysis, only linking the starting point or end point of a shared bicycle trip with rail transit stations. They do not differentiate or compare different travel stages such as the starting point, riding route, and destination, making it difficult to reveal the differentiated impact of the built environment at each stage on riding behavior.

[0007] 2. The path reconstruction accuracy is low, and there is a lack of systematic characterization of the environment in which the cycling path is built.

[0008] Current technologies for processing shared bicycle trajectories only call the shortest path algorithm based on the origin-destination (OD) point, ignoring the actual riding path. This leads to a significant deviation between the reconstructed path and the actual riding path, resulting in errors in the extraction of built environment features along the route. Furthermore, existing methods often neglect the detailed characterization of road structure, street environment, and other features along the actual shared bicycle riding path, factors that directly affect the continuity and safety of the ride and users' willingness to use the service.

[0009] 3. The model lacks directional heterogeneity, making it difficult to accurately depict real needs and influencing mechanisms.

[0010] Existing studies typically treat "inbound connection" and "outbound connection" as homogeneous processes, lacking comparative studies on different connection directions. This leads to biases in predicting the actual connection demand between shared bicycles and rail transit, making it difficult to support refined decision-making for optimizing shared bicycle deployment, renovating the space around stations, and integrated transportation planning.

[0011] 4. Traditional statistical models have limited explanatory power and are unable to reveal the complex nonlinearity and interactive effects of the built environment.

[0012] Existing research on shared bicycle-rail transit connection behavior mostly employs linear regression or traditional statistical models. These methods typically assume a linear relationship between built environment factors and connection behavior, and that the variables are independent of each other. This makes it difficult to characterize the complex nonlinear influence of built environment factors on connection behavior and the mechanisms of multi-factor interactions. Therefore, in areas with high subway passenger flow but still low shared bicycle connection rates, traditional models often fail to effectively explain the reasons for this, and struggle to identify key thresholds and synergistic effects influencing connection behavior. In contrast, interpretable machine learning-based methods can improve prediction accuracy while quantifying the importance ranking, nonlinear influence relationships, and interactions between different built environment factors. This reveals the formation mechanism of connection behavior under the combined influence of multiple factors such as urban location conditions, differences in station passenger flow, and differences in the surrounding built environment, thus overcoming the shortcomings of traditional methods in identifying complex mechanisms. Summary of the Invention

[0013] The purpose of this invention is to provide a shared bicycle track connection analysis method that takes into account the built environment during the travel phase, thereby solving the above-mentioned problems existing in the prior art.

[0014] To achieve the above objectives, this invention provides a method for analyzing shared bicycle rail connections that considers the built environment during the travel phase, comprising the following steps: S1. Obtain shared bicycle order trajectory data within the target area, identify shared bicycle connecting to rail transit travel behavior, perform origin-destination OD mapping on the identified connecting travel, and generate shared bicycle travel OD samples; S2. Reconstruct the riding path from the shared bicycle trajectory data corresponding to the OD sample and restore the path. S3. Extract the corresponding built environment features for different travel stages of shared bicycle connection travel; S4. Based on the characteristics of the built environment at different travel stages, construct prediction models for predicting the probability of shared bicycle usage when connecting to rail transit stations and when connecting to rail transit stations. S5. Interpret and analyze the prediction model, quantify the impact, nonlinearity, and interaction of the built environment characteristics at different travel stages on shared bicycle connection behavior, and output the key influencing factors and their relationships.

[0015] In some embodiments of this application, in S1, identifying the travel behavior of connecting shared bicycles with rail transit includes: Create a 100-meter buffer zone at each entrance and exit of the subway station to filter and analyze parking data of shared bicycles; The travel chain is restructured, and the KD-Tree spatial search algorithm is used to identify travel behavior of shared bicycles connecting to rail transit; The identified connecting trips are mapped from origin to destination (OD) to generate shared bicycle trip OD samples and a shared bicycle connecting trip sample set is constructed.

[0016] In some embodiments of this application, in S2, the cycling path reconstruction and path restoration of the shared bicycle trajectory data corresponding to the OD sample includes: The shared bicycle trajectory data is processed using a coordinate system one method to transform the original GPS coordinates to the target map coordinate system; a cycling road topology map is constructed based on road network data, and the origin-destination points are matched to the corresponding road network nodes to form start-point and end-point nodes. For each OD node pair, combined with the topological attributes of the road network, the path planning service is invoked to generate a set of multiple candidate cycling routes on the cycling road topology map; For each candidate path Calculate the overall path cost Select the most likely cycling route based on the minimum cost criterion. ; Extract the node coordinate sequence on the most likely cycling route, generate the latitude and longitude sequence of the cycling route points, and associate them with the corresponding origin and destination information to construct a connecting travel dataset with a three-stage structure of "origin - cycling route - destination".

[0017] In some embodiments of this application, in S2, for each candidate path Calculate the overall path cost The calculation formula is: ; in, The length of the road edge is calculated using the geometric length of the road polyline. Priority coefficients for cycling facilities are extracted from the road attribute field. The road impedance is obtained through road classification. Traffic interference impedance is obtained by considering intersection density, number of traffic lights, and intersection delay level. , All of these are path weight parameters, which are statistically calibrated using historical cycling travel time and distance distribution data. Select the most likely cycling route based on the minimum cost criterion. : .

[0018] In some embodiments of this application, in S3, the built environment features include: The built environment characteristics of the origin and destination stages of shared bicycle trips are extracted. Extracting the built environment characteristics of street space along the cycling route; Extract station attribute features related to rail transit stations.

[0019] In some embodiments of this application, in step S3, the corresponding built environment features are extracted for different travel stages of shared bicycle shuttle travel, including: Based on POI data and land use data, density characteristics, diversity characteristics, public transport facility proximity characteristics, and destination accessibility characteristics within the buffer zone are calculated. Land use mixing is calculated using the entropy index model, and the expression is: ; in, Let Z be the land use mixing entropy, and Z be the number of land use types. The proportion of land use types within each grid cell; Along the reconstructed cycling path S2, panoramic street view images are acquired at fixed step lengths. After acquiring the street view images, the panoramic images are cropped and projected based on the average visual height and horizontal field of view of the human eye during cycling to generate images from the cyclist's perspective. A pre-trained deep learning model is used to perform semantic segmentation on the street view images to automatically identify several types of pixel regions. Based on the pixel ratio, the street greening, sky openness, and street enclosure from the human eye's perspective are calculated, and the indicators of all sampling points on the path are averaged as the micro-environmental features of this cycling trip. Calculate station passenger flow, subway station proximity to centrality, and the number of connecting bus routes around the station; The calculation is performed using spatial or topological distance between stations, and the formula is as follows: ; in, To ensure the proximity of subway stations to the center, For the station With Station The shortest path distance between them.

[0020] In some embodiments of this application, in S3, the formula for calculating the street greening degree is: ; in, For the greening of roads and streets, This represents the pixel percentage of trees in a roadside street scene. The pixel percentage of shrubs in the street scene; The formula for calculating street enclosure is: ; in, For the degree of street enclosure, The pixel percentage of trees in the street view along the cycling path. The pixel ratio of the street view wall along the cycling path. The pixel ratio of street view buildings along the cycling route. The pixel percentage of street view bushes along the cycling path. The pixel ratio of the streets along the cycling route. This represents the pixel percentage of the pedestrian walkway in the street view of the cycling path.

[0021] In some embodiments of this application, S4 includes: For each OD pair of samples, the sample data is divided into two subsets based on the connecting travel direction: inbound connecting and outbound connecting. A complete feature vector containing origin features, path features and station features is constructed for each subset. The dataset for each scenario is divided into a training set for model training and a test set for model validation in a preset ratio of 8:2. For the two scenarios of entering and exiting the station, gradient boosting decision tree prediction models are constructed respectively. The aforementioned feature vectors are used as inputs, and whether a connection occurs is used as the output label. The models are trained on the training set, and the prediction error is reduced through iterative optimization. A gradient boosting decision tree model is efficiently implemented using the LightGBM framework. The trained prediction model is evaluated and optimized using a test set. Key parameters of the LightGBM model include: 10 leaves per tree, a feature sampling ratio of 0.8, a minimum number of samples per leaf node of 20, and a regression objective function. Evaluation metrics include mean squared error, root mean square error, and coefficient of determination R0.2 .

[0022] In some embodiments of this application, S5 includes: The SHAP framework is applied to interpret the trained prediction model; for the inbound and outbound connection models, the SHAP value of each input feature is calculated separately; the SHAP value is based on the Shapley value principle of cooperative game theory, which quantifies the average marginal contribution of the feature to the model prediction result in all possible feature combinations. The following results are generated based on the SHAP value: Feature Dependency Plot: Used to show the non-linear relationship between a single feature value and its SHAP value, representing the threshold and trend of key built environment factors affecting connection behavior; Interaction analysis: used to identify and visualize the interaction effects between two features.

[0023] In some embodiments of this application, in step S5, the expression for calculating the SHAP value of each input feature is: ; In the formula, Let be the SHAP value of feature j. For the total number of features, For a subset of features that does not contain feature j, This represents the model's output value when using a feature subset S; By comparing the average absolute value of the SHAP values ​​of each feature, the global importance ranking of built environment features at different travel stages to the predicted connection probability is determined.

[0024] The advantages and beneficial effects of this invention compared to the prior art are: 1. This invention, by performing phased modeling of the origin, route, and destination of shared bicycle connecting travel, can more comprehensively and realistically reflect the formation mechanism of integrated travel between shared bicycles and rail transit.

[0025] 2. This invention achieves systematic integration of multi-source built environment characteristics at different travel stages, improving the accuracy of analysis and prediction of shared bicycle connection demand.

[0026] 3. This invention constructs entry and exit models respectively, explains the impact mechanism of entry and exit connections, and provides theoretical support for differentiated planning strategies.

[0027] 4. This invention can identify the impact intensity and threshold range of key built environment factors, quantify the interaction of built environment characteristic indicators, and provide a scientific basis for optimizing the space around subway stations, the layout of connecting facilities, and the formulation of transportation policies.

[0028] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0029] Figure 1 This is a flowchart of a shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in an embodiment of the present invention. Figure 2 This is an illustrative diagram illustrating the shared bicycle-subway connection according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the importance ranking of machine learning inbound OD in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the importance ranking of outbound destinations (OD) in machine learning according to an embodiment of the present invention. Figure 5 This is a non-linear relationship graph of machine learning in an embodiment of the present invention; Figure 6 This is a schematic diagram of the interactive graph results of machine learning in an embodiment of the present invention. Detailed Implementation

[0030] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0031] The core inventive point of this invention lies in its pioneering approach to analyzing the built environment impact of shared bicycle and rail transit connections from the perspective of the entire travel process. This method, which analyzes the "origin—cycling path—destination" multi-stage process, breaks through the limitations of existing research paradigms that only consider subway stations or single endpoints as analytical units. Specifically, this invention does not simply attribute shared bicycle connection behavior to the environment surrounding subway stations. Instead, it deconstructs a single connection trip into three interconnected but differently influencing spatial stages: the origin stage, the cycling path stage, and the destination stage. It then quantifies the built environment characteristics corresponding to each stage, thereby achieving a refined characterization of the connection trip decision-making process. The key improvements and the technical problems solved are as follows: 1. Construction of phased travel samples and reconstruction of cycling routes.

[0032] Technical solution: To support the above multi-stage analysis framework, this invention proposes a cycling path reconstruction method based on candidate path generation and comprehensive cost screening. In the absence of complete real cycling trajectories, the most likely cycling path is reconstructed through road topology constraints, cycling facility attributes, and multidimensional impedance functions, so that the built environment exposure of the path stage can be quantified and incorporated into the model analysis.

[0033] Technical problem solved: It overcomes the limitation of existing methods that only analyze stations as units, and achieves a detailed depiction of the complete travel stage of "starting point - cycling route - destination", laying a data foundation for subsequent stage feature extraction.

[0034] Advantages: It can distinguish and compare the impact of the built environment at different stages of travel, more realistically reflect the entire process of cycling-riding travel, and improve the comprehensiveness and accuracy of the analysis.

[0035] 2. Construct a multi-stage, multi-dimensional environmental characteristic system.

[0036] Technical Solution: A systematic approach is used to extract multi-source built environment features for the starting point, cycling route, and rail transit stations. For the starting point and destination (not just one end of a subway station), traditional built environment indicators such as density, land use mix, public transport facilities, and destination accessibility are emphasized. For cycling routes, street view image data is used to extract street environment features such as road density, sidewalk density, street greening, sky visibility, and spatial enclosure. At the rail transit station level, station attribute indicators such as subway station passenger flow, station network centrality, and connecting bus routes are extracted.

[0037] Technical problem to be solved: Existing technologies fail to distinguish the environmental impacts of different travel stages, resulting in the mixed processing of various built environment factors, making it difficult to reveal their differentiated mechanisms of action.

[0038] Corresponding advantages: By extracting environmental features in stages, a systematic characterization of the impact of the built environment at different stages from the starting point to the path to the destination can be achieved, thereby improving the refinement and interpretability of the analysis results.

[0039] 3. Machine learning-based connection probability prediction and interaction analysis.

[0040] Technical Solution: For both inbound and outbound connection scenarios, the LightGBM machine learning model is employed. Using built-up environment features from the origin, route, and destination as input, it predicts the probability of using shared bicycles for rail connections. The SHAP framework is further utilized to interpret the model, quantifying the global importance, local contribution, and nonlinear impact of each built-up environment feature, and identifying the interaction effects between key features.

[0041] Technical problems solved: It solved the problem that traditional methods could not accurately quantify the complex and nonlinear impact of built environment factors on connection behavior; through the interaction between built environment indicators, it explained why the connection rate is still low in some areas with high subway passenger flow (such as different distances from the city center), which is something that traditional statistical models cannot discover.

[0042] Corresponding advantages: It not only achieves high-precision forecasting of connection demand, but also reveals the direction, intensity and threshold of the effects of various environmental factors, providing a clear basis for optimizing the environment around subway stations, configuring shared bicycle facilities and formulating connection strategies.

[0043] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0044] like Figure 1 As shown, this invention provides a method for analyzing shared bicycle rail connections that considers the built environment during the travel phase, including the following steps: S1. Obtain shared bicycle order trajectory data within the target area, identify shared bicycle connecting to rail transit travel behavior, perform origin-destination (OD) mapping on the identified connecting trips, and generate shared bicycle travel OD samples.

[0045] S2. Reconstruct the riding path from the shared bicycle trajectory data corresponding to the OD sample and restore the path.

[0046] S3. Extract the corresponding built environment features for different travel stages of shared bicycle connection travel.

[0047] S4. Based on the built environment characteristics of different travel stages, construct prediction models for predicting the probability of shared bicycle usage for rail transit entry (riding towards the subway station) and exit (riding out of the subway station).

[0048] S5. Interpret and analyze the prediction model, quantify the impact, nonlinearity, and interaction of the built environment characteristics at different travel stages on shared bicycle connection behavior, and output the key influencing factors and their relationships.

[0049] This invention, by performing phased modeling of the origin, route, and destination of shared bicycle connecting trips, can more comprehensively and realistically reflect the formation mechanism of integrated travel between shared bicycles and rail transit; this invention achieves systematic integration of multi-source built environment characteristics at different travel stages, improving the accuracy of analysis and prediction of shared bicycle connecting demand.

[0050] In some embodiments of this application, in S1, identifying the travel behavior of connecting shared bicycles with rail transit includes: Create a 100-meter buffer zone at each entrance and exit of the subway station to filter and analyze parking data of shared bicycles; The travel chain is restructured, and the KD-Tree spatial search algorithm is used to identify travel behavior of shared bicycles connecting to rail transit; The identified connecting trips are mapped from origin to destination (OD) to generate shared bicycle trip OD samples and a shared bicycle connecting trip sample set is constructed.

[0051] In some embodiments of this application, in S2, the cycling path reconstruction and path restoration of the shared bicycle trajectory data corresponding to the OD sample includes: The shared bicycle trajectory data is processed using a coordinate system one method to transform the original GPS coordinates to the target map coordinate system; a cycling road topology map is constructed based on road network data, and the origin-destination points are matched to the corresponding road network nodes to form start-point and end-point nodes. For each OD node pair, combined with the topological attributes of the road network (one-way street, dedicated cycling lane), the route planning service (such as the cycling mode of Gaode API) is invoked to generate a set of multiple candidate cycling routes on the cycling road topology map; For each candidate path Calculate the overall path cost Select the most likely cycling route based on the minimum cost criterion. ; Extract the node coordinate sequence on the most likely cycling route, generate the latitude and longitude sequence of the cycling route points, and associate them with the corresponding origin and destination information to construct a connecting travel dataset with a three-stage structure of "origin - cycling route - destination".

[0052] In one embodiment, the data structure of each record includes the fields shown in Table 1: Table 1

[0053] In some embodiments of this application, in S2, for each candidate path Calculate the overall path cost The calculation formula is: ; in, The length of the road edge is calculated using the geometric length of the road polyline. Priority coefficients for cycling facilities are extracted from the road attribute field. The road impedance is obtained through road classification. Traffic interference impedance is obtained by considering intersection density, number of traffic lights, and intersection delay level. , All of these are path weight parameters, which are statistically calibrated using historical cycling travel time and distance distribution data. Select the most likely cycling route based on the minimum cost criterion. : .

[0054] In some embodiments of this application, in S3, the built environment features include: The built environment characteristics of the origin and destination stages of shared bicycle trips are extracted. Extracting the built environment characteristics of street space along the cycling route; Extract station attribute features related to rail transit stations.

[0055] In some embodiments of this application, in step S3, the corresponding built environment features are extracted for different travel stages of shared bicycle shuttle travel, including: Based on POI data and land use data, density characteristics, diversity characteristics, public transport facility proximity characteristics, and destination accessibility characteristics within the buffer zone are calculated. Land use mixing is calculated using the entropy index model, and the expression is: ; in, Let Z be the land use mixing entropy, and Z be the number of land use types. The proportion of land use types within each grid cell; Along the reconstructed cycling path S2, panoramic street view images are acquired at fixed step lengths (e.g., every 50 meters). After acquiring the street view images, the panoramic images are cropped and projected based on the average eye level (e.g., 1.5 meters) and horizontal field of view (e.g., 120 degrees) of the cyclist to generate images from the cyclist's perspective. A pre-trained deep learning model, such as PSPNet or DeepLabV3+, is used to perform semantic segmentation on the street view images, automatically identifying several types of pixel regions. The street greening degree from the cyclist's perspective is calculated based on the pixel ratio. ), openness of the sky ( The data includes street enclosure and the average of all sampling points along the route, which are then used as the micro-environmental characteristics of the ride. Calculate station passenger flow, subway station proximity to centrality, and the number of connecting bus routes around the station; The calculation is performed using spatial or topological distance between stations, and the formula is as follows: ; in, To ensure the proximity of subway stations to the center, For the station With Station The shortest path distance between them.

[0056] In some embodiments of this application, in S3, the formula for calculating the street greening degree is: ; in, For the greening of roads and streets, This represents the pixel percentage of trees in a roadside street scene. The pixel percentage of shrubs in the street scene; The formula for calculating street enclosure is: ; in, For the degree of street enclosure, The pixel percentage of trees in the street view along the cycling path. The pixel ratio of the street view wall along the cycling path. The pixel ratio of street view buildings along the cycling route. The pixel percentage of street view bushes along the cycling path. The pixel ratio of the streets along the cycling route. This represents the pixel percentage of the pedestrian walkway in the street view of the cycling path.

[0057] In some embodiments of this application, S4 includes: For each OD pair of samples, the sample data is divided into two subsets based on the connecting travel direction: inbound connecting and outbound connecting. A complete feature vector containing origin features, path features and station features is constructed for each subset. The dataset for each scenario is divided into a training set for model training and a test set for model validation in a preset ratio of 8:2. For the two scenarios of entering and exiting the station, gradient boosting decision tree prediction models are constructed respectively. The aforementioned feature vectors are used as inputs, and whether a connection occurs is used as the output label. The models are trained on the training set, and the prediction error is reduced through iterative optimization. A gradient boosting decision tree model is efficiently implemented using the LightGBM framework. The trained prediction model is evaluated and optimized using a test set. Key parameters of the LightGBM model include: 10 leaves per tree (num_leaves), a feature sampling ratio (feature_fraction) of 0.8, a minimum number of samples per leaf node (min_child_samples) of 20, and regression as the objective function. Evaluation metrics include mean squared error, root mean square error, and coefficient of determination R0. 2 .

[0058] The core objective function of the LightGBM framework is shown below: ; in, To measure the predicted value With real labels The loss function for the difference between them; This is a regularization term used to control all decision trees. To reduce complexity and prevent overfitting; In some embodiments of this application, S5 includes: The SHAP framework is applied to interpret the trained prediction model; for the inbound and outbound connection models, the SHAP value of each input feature is calculated separately; the SHAP value is based on the Shapley value principle of cooperative game theory, which quantifies the average marginal contribution of the feature to the model prediction result in all possible feature combinations. The following results are generated based on the SHAP value: (a) Feature Dependency Plot: Shows the non-linear relationship between a single feature value and its SHAP value (i.e. the direction and magnitude of its influence on the prediction), revealing the threshold and trend of key built environment factors (such as street greening rate and land use mixing) affecting connection behavior; (b) Interaction analysis: Identify and visualize the interaction effects between two features, i.e. how the influence of one feature on the prediction changes with the change of the value of another feature, thereby deeply exploring the synergistic or antagonistic mechanisms between features of the multi-stage built environment and solving the endogeneity problem.

[0059] In some embodiments of this application, in step S5, the expression for calculating the SHAP value of each input feature is: In the formula, Let be the SHAP value of feature j. For the total number of features, For a subset of features that does not contain feature j, This represents the model's output value when using a feature subset S; By comparing the average absolute value of the SHAP values ​​of each feature, the global importance ranking of built environment features at different travel stages to the predicted connection probability is determined.

[0060] In some embodiments, the shared bicycle-subway connection diagram of the present invention is as follows: Figure 2 As shown in the diagram, the importance ranking of machine learning inbound OD in this invention is illustrated in the following figure. Figure 3 The importance ranking diagram of outbound OD in machine learning in this invention is shown below. Figure 4 The nonlinear relationship graph of machine learning in this invention is as follows: Figure 5 The interactive graph results of machine learning in this invention are illustrated in the following diagram. Figure 6 .

[0061] Figure 3 and Figure 4 The variable names are explained as follows: Population_density; Employment_density; Land_use_mixture; Jobs-housing_ratio; Commercial_facilities; Transportation_facilities; Educational_facilities; Distance to_city_center; Metro_ridership_inbound; Closeness_centrality; Feeder_bus_lines; Green_view_index; Sky_view_index; Enclosure; Sidewalk_index; Road_index; Cycling_distance.

[0062] In this application, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. In case of any inconsistency, the meaning set forth in this specification or derived from the content described herein shall prevail. Furthermore, the terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit the scope of this application.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for analyzing shared bicycle rail connections considering the built environment during travel phases, characterized in that, Includes the following steps: S1. Obtain shared bicycle order trajectory data within the target area, identify shared bicycle connecting to rail transit travel behavior, perform origin-destination OD mapping on the identified connecting travel, and generate shared bicycle travel OD samples; S2. Reconstruct the riding path from the shared bicycle trajectory data corresponding to the OD sample and restore the path. S3. Extract the corresponding built environment features for different travel stages of shared bicycle connection travel; S4. Based on the characteristics of the built environment at different travel stages, construct prediction models for predicting the probability of shared bicycle usage when connecting to rail transit stations and when connecting to rail transit stations. S5. Interpret and analyze the prediction model, quantify the impact, nonlinearity, and interaction of the built environment characteristics at different travel stages on shared bicycle connection behavior, and output the key influencing factors and their relationships.

2. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 1, is characterized in that... In step S1, identifying travel behavior involving shared bicycles connecting to rail transit includes: Create a 100-meter buffer zone at each entrance and exit of the subway station to filter and analyze parking data of shared bicycles; The travel chain is restructured, and the KD-Tree spatial search algorithm is used to identify travel behavior of shared bicycles connecting to rail transit; The identified connecting trips are mapped from origin to destination (OD) to generate shared bicycle trip OD samples and a shared bicycle connecting trip sample set is constructed.

3. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 2, is characterized in that... In step S2, the cycling path reconstruction is performed on the shared bicycle trajectory data corresponding to the OD sample. The path restoration includes: The shared bicycle trajectory data is processed using a coordinate system one method to transform the original GPS coordinates to the target map coordinate system; a cycling road topology map is constructed based on road network data, and the origin-destination points are matched to the corresponding road network nodes to form start-point and end-point nodes. For each OD node pair, combined with the topological attributes of the road network, the path planning service is invoked to generate a set of multiple candidate cycling routes on the cycling road topology map; For each candidate path Calculate the overall path cost Select the most likely cycling route based on the minimum cost criterion. ; Extract the node coordinate sequence on the most likely cycling route, generate the latitude and longitude sequence of the cycling route points, and associate them with the corresponding origin and destination information to construct a connecting travel dataset with a three-stage structure of "origin - cycling route - destination".

4. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 3, is characterized in that... In S2, for each candidate path Calculate the overall path cost The calculation formula is: ； in, The length of the road edge is calculated using the geometric length of the road polyline. Priority coefficients for cycling facilities are extracted from the road attribute field. The road impedance is obtained through road classification. Traffic interference impedance is obtained by considering intersection density, number of traffic lights, and intersection delay level. , All of these are path weight parameters, which are statistically calibrated using historical cycling travel time and distance distribution data. Select the most likely cycling route based on the minimum cost criterion. : 。 5. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 4, is characterized in that... In S3, the built environment characteristics include: The built environment characteristics of the origin and destination stages of shared bicycle trips are extracted. Extracting the built environment characteristics of street space along the cycling route; Extract station attribute features related to rail transit stations.

6. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 5, is characterized in that... In S3, the corresponding built environment features are extracted for different travel stages of shared bicycle shuttle travel, including: Based on POI data and land use data, density characteristics, diversity characteristics, public transport facility proximity characteristics, and destination accessibility characteristics within the buffer zone are calculated. Land use mixing is calculated using the entropy index model, and the expression is: ； in, Let Z be the land use mixing entropy, and Z be the number of land use types. The proportion of land use types within each grid cell; Along the reconstructed cycling path S2, panoramic street view images are acquired at fixed step lengths. After acquiring the street view images, the panoramic images are cropped and projected based on the average visual height and horizontal field of view of the human eye during cycling to generate images from the cyclist's perspective. A pre-trained deep learning model is used to perform semantic segmentation on the street view images to automatically identify several types of pixel regions. Based on the pixel ratio, the street greening, sky openness, and street enclosure from the human eye's perspective are calculated, and the indicators of all sampling points on the path are averaged as the micro-environmental features of this cycling trip. Calculate station passenger flow, subway station proximity to centrality, and the number of connecting bus routes around the station; The calculation is performed using spatial or topological distance between stations, and the formula is as follows: ； in, To ensure the proximity of subway stations to the center, For the station With Station The shortest path distance between them.

7. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 6, is characterized in that... In S3, the formula for calculating the street greening rate is: ； in, For the greening of roads and streets, This represents the pixel percentage of trees in a roadside street scene. The pixel percentage of shrubs in the street scene; The formula for calculating street enclosure is: ； in, For the degree of street enclosure, The pixel percentage of trees in the street view along the cycling path. The pixel ratio of the street view wall along the cycling path. The pixel ratio of street view buildings along the cycling route. The pixel percentage of street view bushes along the cycling path. The pixel ratio of the streets along the cycling route. This represents the pixel percentage of the pedestrian walkway in the street view of the cycling path.

8. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 7, is characterized in that... S4 includes: For each OD pair of samples, the sample data is divided into two subsets based on the connecting travel direction: inbound connecting and outbound connecting. A complete feature vector containing origin features, path features and station features is constructed for each subset. The dataset for each scenario is divided into a training set for model training and a test set for model validation in a preset ratio of 8:

2. For the two scenarios of entering and exiting the station, gradient boosting decision tree prediction models are constructed respectively. The aforementioned feature vectors are used as inputs, and whether a connection occurs is used as the output label. The models are trained on the training set, and the prediction error is reduced through iterative optimization. The gradient boosting decision tree model is efficiently implemented using the LightGBM framework. The trained prediction model is evaluated and optimized using a test set. Key parameters of the LightGBM model include: 10 leaves per tree, a feature sampling ratio of 0.8, a minimum number of samples per leaf node of 20, and a regression objective function. Evaluation metrics include mean squared error, root mean square error, and coefficient of determination R0. 2 .

9. A shared bicycle rail connection analysis method considering the built environment during travel phases, as described in claim 8, is characterized in that... S5 includes: The SHAP framework is applied to interpret the trained prediction model; for the inbound and outbound connection models, the SHAP value of each input feature is calculated; the SHAP value is based on the Shapley value principle of cooperative game theory, which quantifies the average marginal contribution of the feature to the model prediction result in all possible feature combinations. The following results are generated based on the SHAP value: Feature Dependency Plot: Used to show the non-linear relationship between a single feature value and its SHAP value, representing the threshold and trend of key built environment factors affecting connection behavior; Interaction analysis: used to identify and visualize the interaction effects between two features.

10. The shared bicycle rail connection analysis method considering the built environment during the travel phase, as described in claim 9, is characterized in that... In step S5, the expression for calculating the SHAP value of each input feature is as follows: ； In the formula, Let be the SHAP value of feature j. For the total number of features, For a subset of features that does not contain feature j, This represents the model's output value when using a feature subset S; By comparing the average absolute value of the SHAP values ​​of each feature, the global importance ranking of built environment features at different travel stages to the predicted connection probability is determined.

Images