A Taxicab Strong Coverage Station Location Model Based on Big Data of Travel Trajectory ()
1. Introduction
With the continuous development of China’s social economy, the level of road traffic development in China has been significantly improved. In the context of the total length of the country’s roads and the continuous growth of the country’s motor vehicle ownership, traffic congestion has become a common problem plaguing Chinese cities. As an important part of the urban transportation system, a reasonable number of well-distributed taxi strong coverage stops can provide taxi drivers with efficient operational efficiency, rational use of taxi resources, effectively reduce malicious competition in the industry, and also provide passengers with a fast and comfortable waiting experience.
The current urban design codes implemented in my country all have requirements for setting up taxi stops on road sections with high passenger flow, but they have not given specific planning methods. In the actual planning process, less consideration is given to the travel characteristics and needs of passengers and taxis, which can easily cause unreasonable stop settings, which not only affects the utilization rate of taxi stops, but also is not conducive to convenient travel for passengers. In recent years, as various advanced technologies such as GPS and Beidou satellite navigation technology have been widely used in the transportation industry, many domestic scholars have conducted research on the location of taxi stops. By analyzing the temporal and spatial aggregation characteristics of taxi GPS data, the peak time and hot spots of taxi travel demand were obtained [1] [2] [3] [4] . Ye Zhen, Khare R and others analyzed the coordinated relationship between travel demand and urban morphology, and established a taxi stop location model [5] [6] [7] . You J et al. proposed a dual-target location decision-making method for taxi stops based on GPS trajectory data [8] [9] . Meng PC and others apply the methods of overall planning and multiple regression analysis to establish an optimal mathematical model of the degree of matching between supply and demand [10] [11] . Feyereisen T L proposed four major location areas and taxi stops to meet the functional needs of taxi stops [12] . Yasuo et al. introduced the actual living and travel conditions of residents in the site selection task as a reference [13] [14] . Song L et al. established a taxi picking point location model in order to match the location of taxi picking points more closely with the needs of taxis [15] .
At present, the majority of urban taxi traffic research and management in China obtains data from manual survey data and IC card record data. Therefore, it requires huge workload investment in the early stage, and the study of time and location cannot be fully covered, and the data is difficult to obtain and accuracy cannot be guaranteed. In this paper, through the analysis and research of taxi operation data of a typical working day in Nanjing, we can grasp the current situation and characteristics of its operation scientifically and reasonably, and get the strong coverage of taxi stopping points in Nanjing, so as to provide a basis for the transportation department to make decisions and people to travel.
2. Materials and Methods
2.1. Overview of the Study Area
In this paper, Nanjing, Jiangsu Province, is selected as the study area. Nanjing, as the capital city of Jiangsu Province, is geographically located at an important node where the eastern coast meets the Yangtze River basin, with a central location of 118˚46'41''E, 32˚03'26''N, and a total area of 6587 square kilometers. As of 2019, the total road mileage in Nanjing reached 10,182 km, the resident population reached 8.5 million people, and public transportation in the city includes public buses and trams, rental cars and rail transportation, of which 12,083 vehicles are operated by rental cars, and the total number of passenger trips reached 11,442,000. The extremely high volume of people during peak periods poses a serious challenge to local traffic conditions, with most major arterial roads blocked by traffic. The large number of private cars that currently exist in Nanjing carries a small proportion of the traffic volume. In order to rationalize the use of public transport resources, public transport needs to be encouraged, and for this reason it is necessary to provide strong taxi coverage stops in Nanjing.
2.2. Data Acquisition and Preprocessing
2.2.1. Data Acquisition
The data for this study was obtained through the big data trading platform-Data Hall, and all the data counted were 3,242,557 GPS track records, containing 1000 taxis, and the size of the whole data text obtained was about 0.6 GB. The results were obtained by selecting a typical working day (September 10, 2010, 0:00 - 23:59, Friday) of Nanjing city taxi data for statistical analysis to obtain the results.
Taxi data records the operating status of the taxi, and records the real-time operating status of the taxi in detail. The attributes of the original data obtained by the taxi include information such as positioning date, latitude and longitude coordinates, speed, direction angle, operating status, etc., as shown in Table 1.
The road network data used in this paper is the Nanjing city road network data cropped from OSM China mainland data. The original coordinate system is WGS_1984, which is transformed to Beijing 1954 coordinate system by projection. The analysis of selected data from all rental car data reveals that the distribution of rental cars is dense in Qinhuai, Gulou, Xuanwu and Jiangning districts, etc. In order to facilitate processing and research, the road network of the study area is cropped as shown in Figure 1.
2.2.2. Data Preprocessing
The analysis of the raw data reveals that we need to eliminate duplicate and irrelevant data (such as GPS drift points) from the raw data, and we also need to complete the missing parts of the raw data, which also need to be removed as there are cases of too large (more than 3 hours) or too small (less than 5 minutes) operating hours in the actual operation, which will greatly interfere with the research results. In the urban road network, the travel speed of taxis will not exceed 100 km∙h−1 data, and therefore also need to be removed for travel speeds exceeding 100 km∙h−1 data.
When doing spatial analysis, the data used is a geographic coordinate system and we need to convert the geographic coordinate system into a projection coordinate system. The spherical coordinates are converted to plane coordinates by projection transformation. Use the Project tool in ArcGIS to convert the coordinate system in ArcGIS, you can enter the following parameters in order in the tool interface, and the meaning of the parameters is shown in Table 2.
The basic data is the WGS-84 coordinate system. The Beijing 1954 coordinate system is used in the research process of this article, and the WGS-1984 coordinate system needs to be converted to the Beijing 1954 coordinate system. The parameter settings required for coordinate system conversion are shown in Table 3.
![]()
Figure 1. Part of the road network data in Nanjing.
![]()
Table 2. Explanation of the parameters in the project tool.
![]()
Table 3. WGS_1984 coordinate system to Beijing 1954 coordinate system seven parameter method of each parameter value.
2.3. Research Methods
2.3.1. Technical Route
In this paper, after data pre-processing and coordinate system conversion of the obtained taxi operation data, we write a Python program to extract the idling rate of taxi cars and the travel volume of taxi cars in different time periods, and filter the statistics by time period to get the peak travel time, in order to facilitate the study, we select a certain peak time taxi data in ArcGIS for kernel density analysis to get the travel hotspot area, and finally apply the method of extracting hilltop points to get taxi strong coverage stop candidate points, the technical roadmap is shown in Figure 2.
2.3.2. Extraction of Idling Rate
The empty rate referred to in this paper refers to the situation in which a taxi is not carrying passengers while it is in motion. The empty rate is the ratio of the number of vehicles in the empty state to the total number of vehicles in a trip per unit of time. By writing a Python program to count the total number of taxis traveling in different time periods and then count the number of vehicles in empty state among all the taxis traveling. The main code is shown in Figure 3.
![]()
Figure 3. Extracting the main code of empty driving rate.
2.3.3. Extraction of the Number of Taxi Trips during Different Time Periods
The inconsistency in the time of taxi trips leads to differences in the efficiency of taxi operations during different hours. By writing a Python program to count the number of taxi trips in each hour of the day, the extracted data is imported into Excel to make a graph. The main code of this section is shown in Figure 4.
![]()
Figure 4. Extracting the main codes of taxi trips in different time periods.
2.3.4. Extraction of Travel Hotspot Areas
Select the data for a certain peak travel time period, spread the data according to latitude and longitude in ArcGIS and export the data to shp format, then perform a coordinate system conversion to Beijing 1954 coordinate system.
The Kernel Density Analysis tool in ArcGIS uses a kernel function to calculate the amount of area per unit based on point or line elements to fit each point or line to a smooth conical surface. The point elements can be transformed into a raster map to analyze the travel hotspot areas. In setting the parameters, the search radius will be set to 500 m by referring to the service area of public transport stations in the urban road traffic planning and design specification, where the maximum service radius of taxi stop is 500 m.
The nuclear density is calculated as.
(1)
where K is the kernel density raster value, Z is the recorded point, i and j are the horizontal and vertical distances, and r is the search radius.
2.3.5. Extraction of Taxi Strong Coverage Stop Sites
After obtaining the travel hotspot areas, candidate points for taxi strong coverage stops can be extracted from the kernel density map in analogy to the extraction of hilltop points from the DEM digital elevation model raster map.
Firstly, the kernel density map is analyzed in the neighborhood, and the radius of the analysis area is set as a circle, 500 m, and the maximum value is taken. Using the “raster calculator tool”, the points with the same raster value as the original kernel density map are extracted. The remaining points are the candidate points for the strong taxi coverage stops by eliminating the areas without travel demand.
Neighborhood analysis to find the maximum value is given by
(2)
where M is the required grid value, i and j are the horizontal and vertical distances, and r is the search radius.
The equation to find the candidate points is calculated as
(3)
P is the candidate point for the requested taxi stop.
3. Results and Discussion
3.1. Vacancy Rate
Figure 5 shows the temporal trend of the taxi idling rate in Nanjing on September 10, 2010. The emptying rate was at the peak between 5:00 and 6:00, with an emptying rate of about 74%; the emptying rate dropped abruptly between 6:00 and 13:00, from 74% to about 18%, with a significant drop; the emptying rate was about 22% between 14:00 and 19:00, with a more moderate fluctuation; the emptying rate showed an obvious rising trend after 21:00.
It can be seen that the empty vehicle rate has been low during the commuting time, and the empty vehicle rate has increased significantly before and after this period of time.
![]()
Figure 5. Time-varying curve of empty driving rate and number of taxis.
3.2. Temporal Variability
The number of taxi trips varies at various times of the day, with a clear wave pattern of trips over a 24-hour period, as shown in Figure 6. The number of trips is lowest between 5:00 and 6:00, with only 0.08 million trips∙h−1 The number of trips is lowest during the period 5:00 - 6:00, with only 0.08 million trips-; it increases significantly after 8:00 and stays above 0.2 million trips-until 22:00 h−1 The number of trips is generally higher between 11:00 and 20:00, with the largest peak occurring between 18:00 and 19:00, when the number of taxi trips exceeds 0.35 million∙h−1, with taxi trips showing a clear downward trend after 19:00.
3.3. Travel Hotspot Areas
By analyzing the idling rate and the number of trips at different times of the day, a representative travel hotspot between 18:00 and 19:00 was selected for kernel density analysis. The kernel density analysis tool was used in ArcGIS, and the image size was set to 5 m, and the search radius was 500 m. The hotspots of travel demand in the time period of 18:00 - 19:00 are shown in Figure 7.
3.4. Location of Taxi Strong Coverage Stops
After analysis, the kernel density map of travel demand from 18:00 - 19:00 was selected as the object of study for this step.
First, use the “Domain Analysis Tool” in ArcGIS to extract the maximum value in the analysis window and assign it to the focus, and set the size of the analysis window to 500 m. Then, using the raster calculator, we first select the image elements with a raster value of 0 from the focus statistics map, then calculate the image elements with a raster value of 0 from the focus statistics map,
![]()
Figure 6. Time-varying curve of taxi trip volume.
![]()
Figure 8. Candidate points for taxis with strong coverage.
subtract the two, multiply the result with the original kernel density map and convert it to a point file, and get the candidate points of taxi stops with the value size of travel demand.
The results are sorted by high demand, and the 30 points with the highest travel demand are selected as candidate stops for regular travel hotspot areas, and then the next 30 points are selected as candidate stops for episodic travel hotspot areas. The candidate points for taxi strong coverage stops are shown in Figure 8.
4. Conclusion
This paper analyzes the spatio-temporal aggregation characteristics of taxis based on the big data of taxi travel trajectories, and then extracts the candidate points of taxi strong coverage stops. The experimental results show that the taxi data in the area can reflect the travel demand characteristics of the area, and the travel spatio-temporal aggregation characteristics are more stable, which can provide a basis for the location of the stop. Reasonable taxi stop locations can effectively regulate the taxi operating environment, provide a good waiting environment for passengers and improve the traffic situation in the city. This paper only uses taxi data as the basis for the study, which has certain limitations. For accurate determination of the location of the stopping points, it is necessary to combine with the field and refer to the taxi stopping point setting specification to select the appropriate points.