A private electric bicycle sparse track-oriented travel behavior analysis method

Abstract

The present application relates to the technical field of data analysis, in particular to a kind of travel behavior analysis method for private electric bicycle sparse trajectory, S1, point cluster in trajectory data is identified, point cluster includes potential stay area and determined stay area, and the travel behavior of private electric bicycle is extracted according to the time sequence relationship of point cluster;S2, 14 influence factors and the nonlinear relationship of private electric bicycle travel are established using spatiotemporal random forest method;S3, the classic SHAP method is extended, and Shapley value is mapped to time axis and space axis, and the spatiotemporal variation effect of driving factor is explored.The present application can directly infer the travel behavior of human being through the real-time position of electric bicycle, and the dependence of travel behavior extraction on order data is reduced.The present application also establishes a nonlinear driving mechanism modeling method considering spatial correlation and heterogeneity, not only realizes higher fitting accuracy, but also can explore the spatiotemporal variation effect of driving factor.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method for analyzing travel behavior based on sparse trajectories of private electric bicycles. Background Technology

[0002] Currently, scholars both domestically and internationally have established various analytical methods for electric bicycle travel to explore its driving mechanisms. For example, Wang et al. (2024) established a framework for assessing the rideability of urban streets based on electric bicycle trajectories, quantifying the quantitative linear relationship between riding demand and urban environmental rideability; Zhou et al. (2022) established a method for identifying the commuting needs of urban populations based on electric bicycle trajectories, and used a geographically and temporally weighted regression model (GTWR) to reveal the spatiotemporal impacts of built environment, transportation facilities, socioeconomic conditions, and weather conditions on electric bicycle demand.

[0003] Despite extensive research on electric bicycle travel, shortcomings remain. Existing research primarily focuses on shared electric bicycle travel, with very limited coverage of non-private bicycle travel (Liu et al., 2023; Yu et al., 2022). Secondly, existing methods are limited by the trade-off between fitting ability and interpretability, especially when considering spatiotemporal heterogeneity. For example, linear models can well explain the relationship between electric bicycle travel and driving factors, but their fitting ability is limited (linear regression, geographically weighted regression, spatiotemporally and geographically weighted regression). Nonlinear models have strong fitting ability, but lack interpretability (random forests, support vector machines, etc.).

[0004] In recent years, to strengthen the management of privately owned electric bicycles, some cities have installed monitoring stations to collect their tracks. Unlike the GPS tracks of shared electric bicycles, the tracks collected by monitoring stations exhibit significant sparsity, thus protecting individual privacy. The collection of privately owned electric bicycle tracks has spurred a large amount of research.

[0005] To this end, this application proposes a travel behavior analysis method for sparse trajectories of private electric bicycles, which can reveal phenomena that are difficult to discover using traditional methods, and provides a novel data mining method for gaining a deeper understanding of electric bicycle travel patterns and behavioral characteristics. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention provides a method for analyzing travel behavior based on sparse trajectories of private electric bicycles.

[0007] A method for analyzing travel behavior based on sparse trajectories of private electric bicycles, characterized by the following steps:

[0008] S1, identify point clusters in the trajectory data, including potential stopping areas and confirmed stopping areas, and extract the travel behavior of private electric bicycles based on the temporal relationship of the point clusters;

[0009] S2, an extension of the classic random forest, proposes a spatiotemporal random forest to establish a nonlinear relationship between 14 influencing factors and private electric bicycle travel;

[0010] S3 extends the classic SHAP method by proposing spatiotemporal SHAP, which maps Shapley values ​​to time and space axes to explore the spatiotemporal variation effects of driving factors.

[0011] Furthermore, in order to better realize the present invention, the data in S1 includes private electric bicycle trajectory data, traffic network data, points of interest data, Gaode congestion data, weather data, housing price data, and population data.

[0012] Furthermore, to better implement the present invention, the stopping areas in the trajectory data in S1 are divided into two main categories: determined stopping areas and potential stopping areas. When the electric vehicle is parked near a monitoring post, the real-time trajectory points of the electric vehicle will be uploaded to the database at a certain frequency, and the trajectory points will form a cluster near the monitoring post, i.e., the determined stopping area. When the electric vehicle is not parked near a monitoring post, the real-time trajectory points of the electric vehicle will be lost in the database. When the trajectory points are detected again, the first detected area may be the area where the electric vehicle is parked, i.e., the potential stopping area. For the potential stopping area, only the first or last point of the trajectory segment needs to be extracted. For the determined stopping area, it is extracted using a spatiotemporal clustering method. When identifying the stopping area in the trajectory points, the electric bicycle's travel can be generated based on the temporal relationship between the stopping areas, including from determined stopping area to determined stopping area, from determined stopping area to potential stopping area, from potential stopping area to determined stopping area, and from potential stopping area to potential stopping area.

[0013] Furthermore, to better realize the present invention, the spatiotemporal random forest method is as follows: The spatiotemporal random forest introduces a spatiotemporal neighborhood on the basis of the classic random forest, and uses internal and external driving factors in the spatiotemporal neighborhood to jointly infer future travel frequencies. The internal driving factors refer to the historical travel frequencies in the target grid, while the external driving factors refer to the urban built environment, spatial location, and time window information in the target grid. The spatiotemporal random forest model further enhances its ability to capture spatiotemporal relationships by integrating external and internal driving factors, thereby improving the modeling accuracy of urban travel. This method needs to fuse the driving factors of the target grid itself and the driving factors in its neighborhood to ensure a one-to-one correspondence between the driving factors and travel frequencies. The specific fusion method is as follows:

[0014]

[0015] In the formula: and Let i represent the k-th driving factor before and after fusion of the target grid i, respectively, where the first driving factor is... Figure 12 The Historical Trip Frequency, the fourteenth driving factor is Figure 12 House Price in Ω i Represents the set of spatiotemporal neighbors of target grid i; Let i represent the distance between the target grid i and the neighboring grid j. According to the above formula, for the 5th to 13th driving factors, the fused driving factor is the sum of the driving factors in the neighborhood; for the 14th driving factor, the fused driving factor is the weighted average of the driving factors in the neighborhood.

[0016] Furthermore, to better implement the present invention, step S3 specifically involves: constructing a three-dimensional cube with the spatial grid index as the S-axis, the time window as the T-axis, and the driving factor as the F-axis; calculating the Shapley value of any driving factor in the three-dimensional cube based on game theory principles; and assigning the Shapley value to the corresponding spatial location and time window through spatial and temporal mapping; after all driving factors in the three-dimensional cube have been mapped, the changes of driving factors with different time periods and different spatial locations can be analyzed; the Shapley value of the driving factor in the three-dimensional cube is calculated by the following formula:

[0017]

[0018] In the formula: STRF represents the spatiotemporal random forest model; Let N represent the k-th driving factor of target grid i; N represents the time window × the total number of spatial grids, which can be decomposed into Ni. S ×NT ; The driving factor in a three-dimensional cube is represented by... Decomposed to obtain; This represents the Shapley value of the driving factor for the s-th grid in the t-th time window; when A value greater than 0 indicates that the driving factor has a positive effect on electric vehicle travel in the s-th grid and the t-th time window; when... When the value is less than 0, it indicates that the driving factor has a negative effect on electric vehicle travel in the s-th grid and the t-th time window.

[0019] The beneficial effects of this invention are:

[0020] Compared to existing methods for trip analysis based on shared electric bicycle trajectories, the method of this invention has better universality. First, this method can directly infer human travel behavior from the real-time location of electric bicycles, reducing the reliance on order data for trip behavior extraction and facilitating its widespread adoption in multiple cities. Second, this invention also establishes a nonlinear driving mechanism modeling method that considers spatial correlation and heterogeneity. Compared to existing modeling methods, this invention not only achieves higher fitting accuracy but also explores the spatiotemporal variation effects of driving factors. Finally, this invention can explore the spatiotemporal variation effects of driving factors within a nonlinear model. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the research area in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the research framework for analyzing the spatiotemporal patterns of electric bicycle travel in this invention;

[0023] Figure 3 The diagram illustrates the extraction of electric bicycle trips in this invention: (a) from a defined stop point to a defined stop point, (b) from a defined stop point to a potential stop point, (c) from a potential stop point to a defined stop point, and (d) from a potential stop point to a potential stop point.

[0024] Figure 4 This is a schematic diagram of the spatiotemporal random forest of the present invention;

[0025] Figure 5 This is a spatial and temporal mapping diagram of the Shapley values ​​of this invention;

[0026] Figure 6 The following is a statistical distribution chart of travel distance and travel duration at different times according to the present invention: (a) travel distance on weekdays, (b) travel distance on non-weekdays, (c) travel duration on weekdays, and (d) travel duration on non-weekdays.

[0027] Figure 7 This is a relative travel frequency diagram between weekdays and non-weekdays for this invention;

[0028] Figure 8 The spatiotemporal distribution diagram of travel frequency according to the present invention is shown below; (a) spatial distribution of travel frequency on weekdays, (b) spatial distribution of travel frequency on non-weekdays, and (c) temporal distribution of travel frequency on weekdays and non-weekdays.

[0029] Figure 9 The following are comparison results of the spatiotemporal random forest and the baseline of the present invention: (a) linear regression, (b) XGBoost, (c) random forest, (d) spatial random forest, (e) temporal random forest, and (f) spatiotemporal random forest.

[0030] Figure 10 The diagram shows the influence weights of the driving factors of this invention and their temporal heterogeneity; (a) the SHAP values ​​of each driving factor; (b) the SHAP values ​​of the density of commercial POIs over time.

[0031] Figure 11 The diagram shows the influence weights of the driving factors and their spatial heterogeneity in this invention; (a) the SHAP values ​​of all driving factors, (b) the number of commercial POIs in the spatial dimension, and (c) the SHAP values ​​of the density of commercial POIs in the spatial dimension.

[0032] Figure 12 This is the driving factor table in the spatiotemporal random forest of this invention;

[0033] Figure 13 This is a table showing the fitting results (mean ± standard deviation) of the spatiotemporal random forest to the baseline in this invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0035] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0036] Figures 1-13 This is a specific embodiment of the present invention, which is a method for analyzing travel behavior based on sparse trajectories of private electric bicycles.

[0037] like Figure 1 As shown, this embodiment takes the area within the Third Ring Road of Wuhan as the study area and divides the study area into a regular grid of 200 meters × 200 meters.

[0038] The dataset used in this embodiment mainly includes private electric bicycle trajectory data, traffic network data, points of interest data, Gaode congestion data, weather data, housing price data, and population data.

[0039] The trajectory data of private electric bicycles was collected through monitoring stakes in the study area. When a private electric bicycle moved into the detection range of a monitoring stake, the real-time location information of the rider was uploaded to the database. The dataset used in this embodiment spans from December 28, 2020 to January 5, 2021, a total of 9 days.

[0040] In addition to private electric bicycle trajectory data, this embodiment utilizes web crawling technology to obtain concurrent traffic network data, point-of-interest data, Gaode congestion data, housing price data, and population data, thereby constructing the driving factors for electric bicycle travel (which will be discussed further below). The traffic network data, point-of-interest data, and Gaode congestion data are sourced from the Gaode Open Platform, the housing price data from Fang.com, and the population data from WorldPop grid.

[0041] like Figure 2 As shown, the travel behavior analysis framework proposed in this embodiment is divided into three parts: private electric bicycle travel extraction based on sparse trajectories, travel mechanism modeling in the spatiotemporal domain, and driving factor analysis considering spatiotemporal heterogeneity. First, this embodiment identifies potential / definite stopping areas in sparse trajectories and extracts the travel behavior of private electric bicycles based on the temporal relationships of these stopping areas. Second, this embodiment extends the classic random forest model by proposing a spatiotemporal random forest method, which establishes a nonlinear relationship between driving factors and electric bicycle travel characteristics in the spatiotemporal domain. Finally, this embodiment extends the classic SHAP method by mapping Shapley values ​​to time and space axes to explore the spatiotemporal changes of driving factors.

[0042] Extracting information about electric vehicle trips is a crucial step in the travel behavior analysis framework. Since private electric bicycles are only tracked near monitoring stations, their trajectories are sparse. Furthermore, unlike shared electric bicycle trajectories, private electric bicycle trajectories lack order data to aid in travel behavior extraction. In light of these differences, this embodiment proposes a method for extracting travel behavior based on private electric bicycle trajectories. Specifically, this embodiment first identifies the stopping areas in the trajectory data, and then generates travel behavior data for private electric bicycles based on the temporal relationships between these stopping areas.

[0043] like Figure 3 As shown, the stopping areas in the trajectory data are mainly divided into two categories: determined stopping areas and potential stopping areas. When an electric vehicle is parked near a monitoring station, its real-time trajectory points will be uploaded to the database at a certain frequency, and the trajectory points will form a cluster near the monitoring station, i.e., a determined stopping area. When the electric vehicle is not parked near a monitoring station, its real-time trajectory points will be lost in the database. When the trajectory points are detected again, the first detected area may be the area where the electric vehicle was parked, i.e., a potential stopping area. For potential stopping areas, this embodiment only needs to extract the first or last point of the trajectory segment. For determined stopping areas, this embodiment extracts them using a spatiotemporal clustering method. When identifying stopping areas in the trajectory points, the electric bicycle's journey can be generated based on the temporal relationship between the stopping areas, including the journey from a determined stopping area to a determined stopping area (…). Figure 3 a) From identified areas of stay to potential areas of stay Figure 3 b) From potential stay area to confirmed stay area Figure 3 c) From potential stay area to potential stay area ( Figure 3 d).

[0044] Random forest models are a traditional machine learning model, and their powerful nonlinear fitting capabilities have led to their successful application in modeling the driving mechanisms of urban travel. While random forest models have achieved satisfactory modeling accuracy, they still have limitations. Specifically, current modeling methods often simply apply random forest models without considering the spatiotemporal relationships inherent in travel data. The classic random forest model's assumption of independent and identically distributed samples is insufficient to adequately model the complex underlying mechanisms and effects of travel data. Therefore, this embodiment extends the classic random forest model by proposing a spatiotemporal random forest method.

[0045] like Figure 4 As shown, Spatiotemporal Random Forest introduces a spatiotemporal neighborhood into the classic random forest model, utilizing both internal and external driving factors within this neighborhood to infer future travel frequencies. Internal driving factors refer to historical travel frequencies within the target grid, while external driving factors include information such as the urban built environment, spatial location, and time window within the target grid. By integrating both internal and external driving factors, the Spatiotemporal Random Forest model further enhances its ability to capture spatiotemporal relationships, thereby improving the accuracy of urban travel modeling. Figure 12 This paper details the 14 driving factors in a spatiotemporal random forest. It's important to note that this embodiment requires fusing the driving factors of the target grid itself with those of its neighbors to ensure a one-to-one correspondence between the driving factors and travel frequencies. Specifically, the fusion method is shown in the following formula.

[0046]

[0047] In the formula: and Let i represent the k-th driving factor before and after fusion of the target grid i, respectively, where the first driving factor is... Figure 12 The Historical Trip Frequency, the fourteenth driving factor is Figure 12 House Price in Ω i Represents the set of spatiotemporal neighbors of target grid i; Let i represent the distance between the target grid i and the neighboring grid j. According to the above formula, for the 5th to 13th driving factors, the fused driving factor is the sum of the driving factors in the neighborhood; for the 14th driving factor, the fused driving factor is the weighted average of the driving factors in the neighborhood.

[0048] Although this embodiment establishes a nonlinear relationship between driving factors and electric vehicle travel in the spatiotemporal domain, it remains unclear how these driving factors influence the model's output. SHapley Additive exPlanation (SHAP) is a method for interpreting machine learning model outputs. It helps understand why users understand the model's decisions by measuring the contribution of each feature to the model's output. However, current SHAP methods primarily reveal the influence weights of driving factors on the model's output, without reflecting the variations of driving factors across different time periods and spatial locations. Therefore, this embodiment extends the traditional SHAP method by mapping the SHAP output to time and space axes, thereby revealing the impact of spatiotemporal heterogeneity on driving factors.

[0049] like Figure 5 As shown, this embodiment establishes a three-dimensional cube with the spatial grid index as the S-axis, the time window as the T-axis, and the driving factor as the F-axis. For any driving factor in the three-dimensional cube, this embodiment can calculate the corresponding Shapley value based on game theory principles, and assign the Shapley value to the corresponding spatial location and time window through spatial and temporal mapping. After all driving factors in the three-dimensional cube have been mapped, this embodiment can analyze the changes of driving factors with different time periods and different spatial locations. The Shapley value of the driving factor in the three-dimensional cube can be calculated by the following formula.

[0050]

[0051] In the formula: STRF represents the spatiotemporal random forest model; Let N represent the k-th driving factor of target grid i; N represents the time window × the total number of spatial grids, which can be decomposed into Ni. S×N T ; The driving factor in a three-dimensional cube is represented by... Decomposed to obtain; This represents the Shapley value of the driving factor for the s-th grid in the t-th time window; when A value greater than 0 indicates that the driving factor has a positive effect on electric vehicle travel in the s-th grid and the t-th time window; when... When the value is less than 0, it indicates that the driving factor has a negative effect on electric vehicle travel in the s-th grid and the t-th time window.

[0052] Analysis of experimental results:

[0053] First, from a macro perspective, we explored the differences in travel distance and duration between weekdays and non-weekdays. Then, from a micro perspective, we analyzed the differences in travel frequency across different time and spatial locations.

[0054] Figure 6 The statistical distribution of travel distance and duration on weekdays and non-weekdays is presented. The results show that travel distance exhibits a clear logarithmic distribution on both weekdays and non-weekdays, while travel duration exhibits a clear Hill distribution. Besides the statistical distribution, subtle differences exist in travel distance and duration between weekdays and non-weekdays. Compared to weekday e-bike trips, the travel distance and duration are slightly shorter on non-weekdays. For example, the percentage of trips less than 2 kilometers increased from 59.7% on weekdays to 60.3% on non-weekdays, and the percentage of trips less than 10 minutes increased from 51.2% on weekdays to 52.7% on non-weekdays. Furthermore, compared to existing literature on shared e-bike and shared bicycle trips, the travel distance and duration of private e-bike trips are slightly longer. On weekdays, 48.7% of e-bike trips involved distances between 1 kilometer and 4 kilometers, and 45.8% involved trips lasting between 5 minutes and 20 minutes. On non-working days, 49.1% of electric vehicle trips covered distances between 1 and 4 kilometers, and 45.3% of electric vehicle trips took between 5 and 20 minutes. This indicates that private electric bicycle travel not only solves the first / last mile problem but also serves the daily commuting needs of the population.

[0055] Figure 7The study demonstrates the difference in travel frequency between weekdays and non-weekdays. The results show that travel frequency exhibits a clear bimodal distribution on weekdays, with the morning peak occurring from 7:30 to 9:00 AM and the evening peak from 4:00 to 6:30 PM. Compared to weekdays, non-weekday travel frequency does not show a clear morning / evening peak. These results in travel frequency are consistent with common sense. Furthermore, Figure 8 The study reveals differences in travel frequency across different time and spatial locations at a microscale. The results show that electric bicycle travel exhibits significant heterogeneity across spatial regions and time windows. Spatially, electric bicycle travel hotspots are mainly concentrated in city centers. Temporally, electric bicycle travel hotspots are concentrated during morning and evening rush hours on weekdays, and throughout the entire day on non-weekdays. Figure 7 The spatiotemporal heterogeneity exhibited indirectly proves the necessity of establishing the STRF model.

[0056] like Figure 9 As shown, this embodiment uses linear regression, XGBoost, and random forest models as benchmark experiments to analyze the advantages of the Spatiotemporal Random Forest model in terms of fitting accuracy. The fitting accuracy of the Spatial Random Forest model and the Temporal Random Forest model is also analyzed to demonstrate the rationality of the model design. The results show that the fitting accuracy of the XGBoost and Random Forest models is significantly better than that of linear regression, indicating a non-linear relationship between electric bicycle travel and influencing factors. Secondly, the fitting accuracy of the Spatial Random Forest model and the Temporal Random Forest model is higher than that of the Random Forest model, demonstrating the rationality of introducing temporal or spatial relationships into the classic random forest model. Finally, the fitting accuracy of the Spatiotemporal Random Forest model is better than that of the Spatial Random Forest model and the Temporal Random Forest model, indicating the necessity of simultaneously modeling spatiotemporal relationships. Furthermore, this embodiment analyzes the stability of the fitting accuracy of the Spatiotemporal Random Forest model under different random seeds, and the results are as follows: Figure 13 As shown in the figure, the results demonstrate that the Spatiotemporal Random Forest model not only possesses good prediction accuracy but also good stability, further proving its advantages.

[0057] Figure 10 (a) illustrates the influence weights of driving factors on e-bike trips. The results show that the number of e-bike trips increases with historical trip frequency. Notably, e-bike trip frequency is significantly affected by factors such as commercial POI density, healthcare POI density, and housing prices. Taking commercial POI density as an example, e-bike trip frequency increases with increasing POI density. In addition to analyzing the influence weights of driving factors on e-bike trip frequency, the framework proposed in this embodiment can further analyze the temporal heterogeneity of these influence weights. Figure 10 (b) This paper demonstrates the impact of commercial POI density on e-bike travel at different time windows. The results show that the weight of the impact of commercial POI density on e-bike travel changes continuously over time and correlates with... Figure 7 The frequency of electric bicycle trips in China shows a high degree of consistency. This means that driving factors have a greater impact on electric bicycle trips during peak travel periods than during off-peak periods.

[0058] In addition to the temporal heterogeneity of the influencing weights, this embodiment can further analyze the differences in influencing weights across different spatial locations. It is worth noting that the method proposed in this embodiment can reveal phenomena that are difficult to detect using traditional methods. Taking population density as an example, both high- and low-population-density areas may promote electric bicycle travel, leading to... Figure 11 (a) It is difficult to directly reveal the correlation between population density and e-bike travel. Compared to Figure 11 (a), Figure 11 (b) and Figure 11 (c) Not only can the spatiotemporal patterns of population density factors be discovered, but the distribution of influence weights (Shapely values) at different spatial locations can also be clarified. Furthermore, this embodiment also reveals that influence weights exhibit a clustering pattern at spatial locations. The results indicate that although the spatial location ranking in Figure 10(a) is relatively low, spatial location is extremely important for electric bicycle travel.

[0059] This embodiment establishes a method for analyzing travel behavior based on private electric bicycle trajectories, and verifies its effectiveness using a private electric bicycle trajectory in Wuhan as an example. First, this embodiment identifies point clusters (potential / definite stopping areas) in the trajectory data and extracts the travel behavior of private electric bicycles based on the temporal relationships of these point clusters. Second, this embodiment proposes a spatiotemporal random forest method to establish the nonlinear relationship between 14 influencing factors and private electric bicycle travel. Finally, this embodiment employs an extended traditional SHAP model to reveal the spatiotemporal variation effects of driving factors. The method proposed in this embodiment can reveal phenomena that are difficult to discover using traditional methods, providing a novel data mining approach for a deeper understanding of electric bicycle travel patterns and behavioral characteristics.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing travel behavior based on sparse trajectories of private electric bicycles, characterized in that, Includes the following steps: S1, identify point clusters in the trajectory data, including potential stopping areas and confirmed stopping areas, and extract the travel behavior of private electric bicycles based on the temporal relationship of the point clusters; S2 extends the classic random forest by proposing a spatiotemporal random forest to establish a nonlinear relationship between 14 influencing factors and private electric bicycle travel. The spatiotemporal random forest method introduces a spatiotemporal neighborhood on the basis of the classic random forest, and uses internal and external driving factors in the spatiotemporal neighborhood to infer future travel frequency. The internal driving factors refer to the historical travel frequency in the target grid, and the external driving factors refer to the urban built environment, spatial location, and time window information in the target grid. The spatiotemporal random forest model further improves the model's ability to capture spatiotemporal relationships by integrating external and internal driving factors, thereby improving the modeling accuracy of urban travel. This method requires fusing the driving factors of the target grid itself with those of its neighbors to ensure a one-to-one correspondence between the driving factors and the frequency of trips. The specific fusing method is shown in the following formula: ； In the formula: and Representing the target grid Before and after fusion There are 14 driving factors, the first of which is Historical Trip Frequency and the fourteenth driving factor is House Price. Indicates the target grid The set of spatiotemporal neighbors; Indicates the target grid With neighboring grid The distance; as shown in the above formula, for the 5th to 13th driving factors, the fused driving factor is the sum of the driving factors in the neighborhood; For the 14th driving factor, the fused driving factor is the weighted average of the driving factors in the neighborhood; S3 extends the classic SHAP method to establish a spatiotemporal SHAP method, mapping Shapley values ​​to time and space axes to explore the spatiotemporal variation effects of driving factors. Specifically, S3 involves constructing a three-dimensional cube with the spatial grid index as the S-axis, the time window as the T-axis, and the driving factor as the F-axis. For any driving factor in the three-dimensional cube, the Shapley value is calculated based on game theory principles. The Shapley value is then assigned to the corresponding spatial location and time window through spatial and temporal mapping. After all driving factors in the three-dimensional cube are mapped, the changes in driving factors with different time periods and spatial locations are analyzed. The Shapley value of the driving factor in the three-dimensional cube is calculated using the following formula: ； In the formula: STRF represents the spatiotemporal random forest model; Indicates the target grid The One driving factor; N represents the total number of time windows × spatial grids, decomposed into ; The driving factor in a three-dimensional cube is represented by... Decomposed to obtain; This represents the Shapley value of the driving factor for the s-th grid in the t-th time window; when A value greater than 0 indicates that the driving factor has a positive effect on electric vehicle travel in the s-th grid and the t-th time window; when... When the value is less than 0, it indicates that the driving factor has a negative effect on electric vehicle travel in the s-th grid and the t-th time window.

2. The method for analyzing travel behavior based on sparse trajectories of private electric bicycles according to claim 1, characterized in that: The data in S1 includes private electric bicycle trajectory data, traffic network data, points of interest data, Gaode congestion data, weather data, housing price data, and population data.

3. The method for analyzing travel behavior based on sparse trajectories of private electric bicycles according to claim 1, characterized in that: The stopping areas in the trajectory data in S1 are divided into two categories: one is the determined stopping area and the other is the potential stopping area. When the electric vehicle is parked near the monitoring station, the real-time trajectory points of the electric vehicle will be uploaded to the database at a certain frequency, and the trajectory points will form a cluster of points near the monitoring station, which is the determined stopping area. When an electric vehicle is not parked near a monitoring station, its real-time trajectory points will be lost in the database. When the trajectory points are detected again, the first detected area may be the area where the electric vehicle was parked, i.e., the potential stopping area. For potential stopping areas, only the first or last point of the trajectory segment needs to be extracted. For confirmed stopping areas, spatiotemporal clustering is used to extract them. When identifying stopping areas in the trajectory points, the electric bicycle's travel is generated based on the temporal relationship between the stopping areas, including from confirmed stopping area to confirmed stopping area, from confirmed stopping area to potential stopping area, from potential stopping area to confirmed stopping area, and from potential stopping area to potential stopping area.

Images