How Does the Built Environment Affect Mechanical Parking Space Planning: A Case Study in **’an City

Abstract

Mechanical parking lots and spaces are known as the “energy saver” of urban space because of their small footprint, high efficiency, and environmental protection. However, the location and number of mechanical parking lots and space planning have become an important part of effectively exerting the function of mechanical parking lots. In order to explore the planning problem of mechanical parking lots, this study used the gradient boosting decision tree–Shapley additive explanations (GBDT-SHAPs) to measure the non-linear impact of the urban built environment on the mechanical parking spaces ratio and extract the optimal threshold of key variables. The results show that land use mix and distance to Bell Tower (CBD) are two key variables affecting mechanical parking space planning, and both have a non-linear relationship with the built environment. The threshold values are 0.83 and 7 km. The results will provide urban and transport planners with strategies for planning mechanical parking lots and spaces.

1. Introduction

The problem of parking and traffic operation efficiency caused by a rapid increase in mobility is very serious. Limited land resources and an increase in car ownership have exacerbated the conflict between parking supply and demand [1]. According to the parking survey in **’an in 2022, there are about 3.347 million car parking spaces in the city, and the average parking space is about 0.85, which is lower than the 1.1–1.3 recommended by the national standard (http://zygh.xa.gov.cn/web_files/zygh/file/2023/10/19/202310191543486971221.pdf, accessed on 2 April 2024). Moreover, improper parking supply and layout planning can result in a 30–50% inefficient operation [2], wasting road resources, reducing traffic efficiency, and increasing environmental pollution.

Compared with the traditional flat parking lots, the mechanical parking lots are smarter. Modern mechanical parking systems use the central controller, data collection, analysis, and processing, and, ultimately, generate intelligent monitoring instructions and send them to the drive system, so as to achieve intelligent vehicle flow planning, parking access, and monitoring. The Smart Parking System (SPS), which uses IoT, AI, and multi-agent systems, can better organize parking lots and manage traffic [3]. The core concept of the system is to automatically allocate parking spaces without human intervention. SPS uses IoT technologies and devices (e.g., sensors) to assist drivers. The mobile application allows users to receive real-time parking information about the selected parking area. They can find the nearest available parking space, which can be booked online in advance, thereby reducing traffic congestion, air pollution, and related health risks [4,5]. Moreover, the mechanical parking space has the advantages of saving space, improving parking efficiency, environmental protection, and energy saving. It can reduce the occupation of land, effectively solve the problem of parking space shortage and low parking efficiency, and thus reduce urban traffic congestion and environmental pollution. Therefore, mechanical parking spaces have important advantages and development potential in urban parking management. The proper planning and design of mechanical parking lots and spaces is the key to effectively relieving traffic congestion and reducing environmental impact. Therefore, it is necessary to carry out a reasonable planning layout of mechanical parking spaces and to determine the location and number of mechanical parking spaces by discussing the influence of factors on the construction of mechanical parking spaces, so as to provide theory and application for parking planning and urban planning.

In order to reveal which factors have a greater impact on the planning and construction of mechanical parking lots and mechanical parking spaces, and to explore what built environments are needed to be equipped with mechanical parking lots and how many mechanical parking spaces are needed, so as to provide a theoretical and practical basis for the planning and construction of mechanical parking lots and parking spaces. Traditional linear regression and non-linear regression have constraints on the trends of the model, which may bias the predictive accuracy of the model and the degree of influence of the independent variable on the dependent variable, and thus bias the planning guidance. Therefore, the gradient boosting decision tree (GBDT) method was used to study the non-linear relationship between the mechanical parking ratio and the built environment. It has been shown that there is a non-linear relationship between the built environment and parking demand [1], and that there is a relationship between parking demand and parking supply [6]; therefore, there should be some relationship between the built environment and parking supply. The effect of built environment factors on the mechanical parking space ratio is quantitatively analyzed from global and local perspectives. The main contributions of this study are as follows: on the one hand, there are few studies on mechanical parking planning. Some studies have analyzed the internal organization, control strategy, site selection and cost, mechanical structure [7,8,9,10,11,12], and the design of the device and intelligent system optimization of the single facilities, while there is a lack of regional planning studies that provide theoretical and practical support for the spatial layout planning of mechanical parking. On the other hand, very few studies have addressed the relationship between car parks and the built environment, especially mechanical parking lots, and there is a lack of analysis of the impact of factors.

There are six sections as follows. Section 2 discusses relevant research in the field of parking lots and built environments. Section 3 introduces the models used in this research, including the GBDT and SHAP models. Section 4 presents the study area and data. Section 5 presents the results. Section 6 provides the discussion and conclusion.

2. Related Works

In the field of transport, parking research is becoming increasingly important. The reasonable setting of parking space is an important means to effectively alleviate the problem of parking difficulties and reduce the impact on dynamic traffic flow and environmental pollution. The built environment can influence parking planning by influencing parking demand, which can provide decision guidance for parking planning. Therefore, this paper summarizes two aspects. The first aspect is the analysis of the basic theory and research methods of the parking and built environment; the second aspect is the research method used for an analysis of the relationship between parking and built environment.

2.1. Parking and Built Environment

The 5D built environment indicator system mainly includes density, diversity, design, destination accessibility, and distance to transit [13]. Population density and parking are the sixth and seventh “D” in “7D” [14], which is developed from the “5D” theory. Several research studies have analyzed the relationship between the built environment and travel behavior [15], such as mode choice [16] and parking behavior [17,18,19].

Parking demand is influenced by built environment factors, such as land use [20], location conditions [21], and so on. For parking demand prediction, macro prediction methods mainly include the parking generation rate model, travel attraction model, traffic volume parking demand model, and multiple regression analysis prediction model [22,23]. With the deepening of parking surveys and demand analysis, parking demand prediction models are gradually enriched with survey methods [24] and a consideration of influencing factors such as traffic demand allocation [25] and parking behavior [26,27]. With the development of data collection and storage technology, the observation scale of parking surveys is becoming smaller and smaller, the collection of continuous parking data is becoming easier and easier, and some scholars have proposed the use of time series and machine learning methods for the short-term prediction of parking demand [28]. Parking cost is the most important variable influencing parking demand, followed by location and road density [1]. There is a non-linear influence of the built environment on the demand and shows an obvious threshold effect [12]. Chen has proposed that the construction index (18.92%) and location (15.23%) are the most important built environment factors influencing parking demand [29]. The parking supply is often predicted by parking demand by considering the supply and demand ratio or the supply and demand uncertainty. It is difficult to choose an appropriate supply and demand ratio. Exploring the relationship between the built environment and mechanical parking space might be a new thought for parking supply analysis and prediction.

2.2. The Research Method of the Relationship between Parking and Built Environment

Traditional methods such as Ordinary Least Squares (OLSs) [30], the logit method [1] h, and Geographical Weighted Regression (GWR) [31] have been used to dig out the relationship between parking-related variables and built environment indicator systems. Machine learning, which has higher predictive performance, has been widely used in the transportation field [21,32,33,34]. Due to higher predictive accuracy and operational efficiency, lots of researchers have used the Gradient Boosting Decision Tree (GBDT) model to analyze the influence of the built environment on travel-related variables [33,35,36], or to predict parking-related variables [21,37].

Machine learning algorithms are black box models, i.e., their inner workings are difficult to understand and explain [38]. The interpretability of machine learning has attracted much attention. Partial dependence plots (PDPs) and accumulated local effects (ALE) plots could describe how features affect the prediction of a machine learning model on average. The ALE method has three advantages: Firstly, the graph created by the ALE method is unbiased. As ALE can deal with the correlation between features, the graph produced is not affected by the joint effect. Secondly, ALE is faster than PDP and requires fewer computations than PDP, and thirdly, the interpretation of ALE graphs is very clear and unambiguous. By removing the effects of correlated variables, one can easily interpret the feature variables and model results [39].

One of the difficulties in using machine learning methods is the interpretation of their results [40]. To overcome this problem, Shapley additive explanations (SHAPs), which are based on the classical Shapley values, are used to provide local explanations and give detailed feature importance and attribution at the level of individual observations. This method has brought great potential for understanding machine learning models of spatial data, as each observed value is geographically referenced and allows spatial visualization of any estimated feature attribution. The SHAP value of y_i for sample i is calculated by the sum of the predicted mean value of all samples and the SHAP value of sample x_ij (($f\left(x_{ij}\right)$, which is the individual SHAP value. j is the feature j). When $f\left(x_{ij}\right)$ > 0, it means that the feature enhances the prediction value and plays a positive role, and vice versa [41]). The implementation of local SHAP as SHAP is provided in software such as Python 3.12.3 and R 4.3.3. Machine learning methods such as GBDT, xgboost, and LightGBM can call the SHAP command in the scikit-learn package.

Although the built environment and parking research methods use the advanced GBDT algorithm [29], there is almost no research on mechanical parking and the interpretability of variables and almost no research on the interpretation of local features. To address the unresolved issue in parking planning, the mechanical parking rate was selected as the dependent variable, and the variables based on the “7D” built environment indicator system were used as the independent variable. The GBDT model was used to establish the non-linear relationship to predict the mechanical parking spaces ratio, and then to interpret the predictions from both global and local perspectives and to reveal the contribution of each variable with the SHAP value. This research will provide a theoretical basis for the design and planning of mechanical parking spaces and lots.

3. Materials and Methods

3.1. GBDT Method (Gradient Boosting Decision Tree)

Among machine learning models, the GBDT model is good at representing non-linear relationships and does not require prior assumptions about the relationships between variables. At the same time, it can provide the relative importance of the variables in the results, which is conducive to revealing the results. Moreover, it can deal with the problem of multi-collinearity within variables. GBDT method uses bagging and adding regularization to avoid noise and outlier data, which makes GBDT more robust [36]. Thus, GBDT is now widely used for regression and classification prediction [42].

GBDT is an advanced ensemble learning model based on the CART algorithm. It uses weak classifiers (CART trees) to construct an ensemble model and trains multiple decision tree models through iterations and overlays to achieve the goal of improving prediction accuracy. In each iteration, based on the previous iteration, the loss function (Equation (1)) and the pseudo-residuals (Equation (2)) are calculated. A new decision tree is constructed, and then all the generated decision trees are weighted and fused (Equation (4)) according to the weights of the decision trees (Equation (3)) by gradient descent. The GBDT model combines the decision tree with the idea of the ensemble in an effective way and improves the prediction accuracy of the model. The following steps are usually included.

$$L_i^{\left(t\right)}={(y_i-f^{(t-1)})\left(x_i\right))}^2$$

(1)

$$r_i^{\left(t\right)}=y_i-f^{(t-1)}\left(x_i\right)$$

(2)

$$y^{\left(t\right)}=learning\_rate$$

(3)

$$F_i(x)=\sum _{t=1}^m{y}^{\left(t\right)}{f}^{\left(t\right)}\left(x_i\right)$$

(4)

where $L_i^{\left(t\right)}$ denotes the loss function value of the i^th sample of the t^th decision tree; $r_i^{\left(t\right)}$ denotes the pseudo residual of the i^th sample; $y_i$ is the actual value of the i^th sample, which is the prediction of the t − 1^th decision tree for the i^th sample; $y^{\left(t\right)}$ denotes the weight of the t^th tree, and learning_rate is the learning rate. $F_i\left(x\right)$ denotes the final dependent prediction for the i^th sample, and $f^{\left(t\right)}(x_i)$ denotes the dependent prediction result for the i^th sample of the t^th tree on the observation sample x. The prediction results of all trees are weighted and summed according to the weights to obtain the final prediction result.

For more information, see Li et al. [38].

3.2. Shapley Additive Explanation (SHAP)

The SHAP is an ex-post model interpretation approach, the core idea of which is to compute the marginal contribution of features to the model output, and then to interpret the “black box model” both globally and locally. The SHAP constructs an additive interpretation model, where all features are considered “contributors”. For each prediction sample, the model generates a prediction value, and the SHAP value is the value assigned to each feature in the sample.

In the SHAP interpretation model, the contribution of the feature element i to the predicted value $f\left(x\right)$ is attributed to the Shapley value of the feature element, which is denoted as ${\mathrm{\varnothing }}_i(f,x)$. The formula is [33]:

$${\mathrm{\varnothing }}_i(f,x)=\sum _{S\subseteq N}\frac{\left|S\right|!(M-\left|S\right|-1)!}{M!}\times [f_x(S\cup \left\{i\right\})-f_x(S)]$$

(5)

where N is the set of all features; S is an order-dependent permutation subset of N, and f is the marginal contribution of feature i to the subset S; M is the total number of input features.

4. Study Area and Data

The study area is the center city of **’an, Shannxi province. **’an land space overall planning (2021–2035) (http://zygh.xa.gov.cn/web_files/zygh/file/2022/11/08/202211081852155156437.pdf, accessed on 4 April 2024) defined the central urban area of **’an is about 766 square kilometers, including seven districts. The total number of parking lots is 2506, of which 216 have mechanical parking spaces. In order to explore the influence of built environment factors on mechanical parking spaces and provide a theoretical basis for the construction of mechanical parking lots, this study selected the proportion of the mechanical parking spaces ratio as the dependent variable, and the range of independent variables is [0, 1]. A value of 0 means that all parking spaces are ordinary parking spaces, 0.5 indicates that mechanical parking spaces account for 50% of the total parking spaces, and 1 indicates that all parking spaces are mechanical parking spaces. The distribution of parking lots (yellow dots) and parking lots including mechanical parking spaces (red dots) in the center city of **’an are shown in Figure 1.

The survey data of parking lots in **’an obtained in this study include the basic characteristics of parking lots, such as latitude and longitude coordinates, number of parking spaces and mechanical parking spaces, cost, and architecture type. The analysis of spatial data must take into account the scale effect and the zoning effect. In order to analyze the built environment around the parking lots, it is necessary to choose the appropriate size of the analysis unit. The whole modeling process requires that the spatial analysis unit should not be too large in order to ensure the validity of the measurement of the built environment on the one hand and to highlight the intrinsic variability of the factors of the built environment on the other hand. In reference to the selection of buffer zones in the study on the built environment and motor vehicle travel behavior [34], this paper finally selects the 500 m buffer zone of the parking lots as the research unit. Ten indicators were selected from the “7D” system to construct the independent variable system. Three parking lot characteristic indicators are cost, area, and architecture type.

The open source population density raster data are obtained from the World Population Data website (https://hub.worldpop.org/ (accessed on 20 May 2024)); Point of Interest (POI) data are obtained using the Python program, which is used to access the Gaud API to crawl, and the coordinate transformation plug-in of ArcGIS 10.2 is used to transform the GCJ-02 coordinate system of Gaud into the WGS-84 coordinate system; the road network data intercepts the road network within the scope of the study through the open source map website (OpenStreetMap, OSM) and converts the dual-line road network into a single-line network. The dual-line roads specify the direction of vehicle travel. For an intersection, there is one intersection regardless of the number of lanes. Without this treatment, multiple intersections are counted based on lane crossings.

The Kolmogorov–Smirnov test (K-S test) is a non-parametric hypothesis test in statistics that is used to test whether a single sample follows a certain distribution or whether two samples follow the same distribution. The K-S test can not only assess the degree of fit of the sample data to the theoretical distribution, which can help us to better understand the characteristics and regularity of the data distribution, but also determine whether the distribution of features is the same in the training set and the test set. If the K-S test value is greater than 0.05, the data follow a normal distribution. Through the statistical analysis of the indicators, the K-S test shows that the significance of each variable is greater than 0.05. The sample follows a normal distribution and has good quality. The multi-collinearity test of the explanatory variables is carried out using SPSS 26.0. The variance inflation factor (VIF) of the variables is less than 10. Therefore, there are no strong collinearity variables, and all independent variables can be included in the model.

Table 1. Summary of variables.

Variables	Description	Min	Max	Mean	Std.
Mechanical parking space ratio	The ratio of mechanical parking spaces to total parking spaces	0	1	0.05	0.18
Land use mix	Entropy index of POI land use type in 500 m circular buffer zone of research object	0	0.92	0.76	0.1
Bus stop density	The number of bus stops in the 500 m circular buffer zone of the research object	0	16.56	3.74	2.77
Transit stop density	The number of transit stops in the 500 m circular buffer zone of the research object	0	7.65	1.05	1.74
Network density	The road length in the 500 m circular buffer zone of the research object	0	21.77	7.95	3.40
Intersection density	The number of interactions in the 500 m circular buffer zone of the research object	0	61.11	11.64	9.27
Population density	The population density in the 500 m circular buffer zone of the research object	0	17.16	1.34	2.56
Distance to CBD	The distance from the center point of the buffer zone to CBD (Bell Tower)	0.21	29.51	7.92	4.29
Fee	Parking fee	0	10	2.11	1.48
Area	The parking lot is located in **’an (Inside the Ming City Wall, Ming City Wall-Second Ring, Second Ring-Third Ring, and Third Ring Outside).	1	4	2.94	0.91
Architecture type	1. Hotel; 2. Public; 3. Temporary; 4. Business office; 5. Relics scenic spot; 6. Housing; 7. Comprehensive business; 8. Hospital; 9. Other	1	9	5.76	2.71

gives the summary of the variables.

5. Results

5.1. The Result of the Prediction Model

A total of 2506 buffer areas (π × 500 × 500) are used to construct models for predicting the mechanical parking spaces ratio. Python is used to conduct the prediction using the GBDT model. First, all samples are divided into 80% training and 20% test groups. Then, the GBDT model is fitted and the parameters are selected using the grid search method. The model parameters were determined using the five-fold cross-validation method. The data set was randomly divided into five subsets of equal sample size. Four of them were selected as the training set and the remaining one as the test set. The depth of the tree is 3 and the number of trees is 150. The prediction R² of the GBDT and OLS models are 37% and 11%. The GBDT model achieves a higher prediction accuracy.

5.2. Global Interpretability

The relative importance of the variables obtained in Python is shown in Figure 2. The relative importance of a variable is the predicted contribution of an independent variable to the dependent variable relative to other independent variables. The higher the relative importance, the greater its effect on the dependent variable. Land use mix (46.2%) and distance to CBD (38.2%) (Bell Tower in **’an) are two of the most important variables affecting the mechanical parking spaces ratio. Followed by the variables architecture type (5.4%), network density (4.2%), and fee (3.2%). Chen et al. [29] proposed that proximity to the city center is the second most important factor and that intersection density has the lowest impact on parking demand, which is consistent with our study.

The ALE plots show a non-linear and threshold effect of factors on the MPS ratio. The blue line indicates the value of the independent variable. The denser the blue line, the higher the frequency of that value. Figure 3a shows that the land use mix is positively associated with the MPS ratio. When the land use mix is less than 0.83, there is no obvious variation in the MPS ratio. While it exceeds 0.83, its effects are faced with a sharp increase. The distance to CBD is negatively associated with the MPS ratio, as shown in Figure 3b. It shows a threshold which is about 7 km. When the MPS ratio is less than 7 km, it faces a large slope. When it exceeds 7 km, the influence is minimal. The reason is that the land resources in the city center are scarce, and mechanical parking spaces are needed to improve the level of land use, meet parking demand, and improve turnover efficiency. Different architecture types have different effects on the MPS ratio (Figure 3c). The hotel, public, and temporary areas need more mechanical parking spaces. The network density has an extreme effect (Figure 3d). The higher or lower the density, the greater the impact on the parking rate. The road network in the city center is dense and the shortage of land requires mechanical parking spaces. The road network outside the suburbs is sparse, and some large attractions, such as parks or commercial centers, have a high demand for parking. See from Figure 3e, the influence of bus stops on MPS ratios is positive and the threshold is 5 units/km². The higher the density, the higher the ratio, mainly because the higher the density, the more likely it is to be in the city center, and, therefore, the higher the requirement for the ratio. See from Figure 3f, the influence of fees on MPS ratios is positive and the threshold is 2.5 yuan/h. The higher the cost, the higher the rate. The main reason is that, on the one hand, the parking fee in the city center is high, and it is more likely to build mechanical parking spaces. On the other hand, the cost of a mechanical parking lot is higher than that of an ordinary parking lot. The effects of population density, intersection density, area and transit stop density are close to zero.

The SHAP tree explainer is used to explain the model globally and locally. The order of the influence of the SHAP value on variables is given from large to small in Figure 4. It is almost the same as the importance of the variables obtained from the GBDT simulation. Land use mix, distance to CBD, and architecture type are three variables with large influence and have the best global interpretability to MPS ratio.

5.3. Local Interpretability

The local interpretability of the model shows the prediction contribution of each parking lot buffer with a different spatial coordinate. Two buffers are randomly selected and show the SHAP value of each variable, which is shown in Figure 5. The base value is the same (0.04787), which is the average of all samples. The red bar indicates the positive influence and the blue color means the negative influence. As shown in Figure 5a, the variables with positive SHAP values are population density and distance to CBD. The remaining variables have a negative influence. The total SHAP value is 0.05, which is higher than the mean value. Figure 5b shows another sample of which variables have different influences. These two samples show that the local interpretability of variables is different from the global interpretability, and the local interpretability varies across buffers.

Figure 6 shows the local SHAP values of six different variables. Figure 6a,b,e,f shows variables of the distance to CBD, land use mix, bus stop density, and fee. These four figures show the influence of variables on the MPS ratio has strong spatial aggregation. Figure 6c,d show architecture type and population density. The influence is more dispersed in space. These graduated symbol maps can suggest the MPS ratio that is dominated by that particular variable. The local SHAP values of different variable distributions are also different. This is because different variables have different influences on the model at different spatial positions. These figures show the different contributions of each variable, which can help urban and traffic planners generate tailored strategies for parking lots and mechanical parking spaces planning.

6. Discussion and Conclusions

The non-linear relationship between the mechanical parking spaces ratio (MPS ratio) and the built environment is constructed by using the GBDT model. Moreover, the ALE plot is used to simulate the influence of variables on the mechanical parking spaces ratio and to determine the influence threshold. In addition, the SHAP value is used to reveal the local feature of built environment variables on the MPS ratio. The local feature means the influence of variables on a sample which is a buffer for a parking lot. Due to the different spatial locations of the different buffers, the built environment around them is different, which makes the effect of the factors on the MPS ratio show different effects. Based on these studies, some valuable insights are provided for the planning of mechanical parking lots and spaces. The results make three contributions to the parking analysis. First, the results reveal the non-linear influence of built environment variables on the mechanical parking spaces ratio. Second, the relative importance of variables on the mechanical parking spaces ratio helps planners make the mechanical parking lots and spaces planning. Third, the use of SHAP values to describe the spatially local impacts of the built environment can reveal the spatial variability of variable impacts.

From Figure 2, the land use mix and the distance to CBD play an important role in the planning of mechanical parking spaces and lots. The construction and maintenance costs of the mechanical parking spaces are higher than those of conventional parking spaces. Therefore, the mechanical parking spaces or parking lots in the necessary locations can not only meet the parking demand but also save costs. In this study, it is more necessary to build mechanical parking spaces or parking lots in an area where the land use mix is higher than 0.83 and the distance from the city center is less than 7 km. However, according to the local effect diagram in Figure 6, the construction of mechanical parking spaces or parking lots can also be considered for areas where the variables have a positive effect. For building types, the impact of the construction of mechanical parking spaces or parking lots can be considered when planning parking lots for building types with a higher impact, such as hotels with higher impact.

According to the local effect diagram in Figure 6, the variables in some areas have a positive impact on the mechanical parking spaces ratio; therefore, it is necessary to build a mechanic parking space for the area with a large value of the variable. The variables in some areas have a negative effect on the mechanical parking spaces ratio, which requires the construction of mechanical parking spaces for areas where the built environment variables have lower values. According to the results of the study, shown in Figure 6a, it can be seen that the positive effect of the variable distance to CBD on the construction of mechanical parking is mainly evident in the city center as well as in the periphery. Thus, the closer to the city center, the more mechanical parking spaces are needed to meet the parking demand. However, the ALE graph shows that, although the distance to the CBD has the second highest impact, the impact trend is not very large. This conclusion also reflects that the CBD does not advocate a lot of parking lots, as this will increase parking demand and thus traffic congestion in the city center area. At the same time, the periphery can be a park-and-ride site and provide a reference for the installation of park-and-ride sites. In addition, as shown in Figure 6f, the higher the parking fees, the more mechanical parking spaces there will be; moreover, mainly because of the high construction costs of mechanical parking garages, only high charges can cover normal operating and maintenance costs. This finding can also provide a basis for charging for mechanical parking lots. Above all, planners should make mechanical parking planning schemes more responsive to the built environment of the area and the travel characteristics of residents (the built environment influences the travel characteristics of residents). This allows the mechanical parking system to play its proper role. Thus, the results of this study provide theoretical and practical references for car park planning and parking space allocation.

This study still has some limitations and issues that need to be discussed and solved in the future. As the empirical analyses are based on a single city, whether the impact of built environment factors on parking demand is equally significant in different cities has not been verified. In the future, we can pay attention to the corresponding problems of different cities and different subregions and explore the changes of the influence mechanism in different regions; moreover, we can also pay attention to whether different cities and regions have the same significant conclusions, in order to better serve the urban construction and to improve the travel environment and improve travel satisfaction.