Abstract

With the aggravation of traffic congestion, it is more and more important to find effective solutions to traffic congestion. In order to promote the solution of traffic congestion, a data-based study on road congestion patterns in Chengdu urban area has been carried out. Firstly, the research progress at home and abroad is analyzed, and the research method based on velocity data is determined. Then the velocity data is analyzed by K-Means clustering algorithm. After verification, a 5-level traffic congestion classification table is obtained. Five different road traffic congestion patterns are identified, including very slow, slow, normal, smooth and very smooth. Then, based on the classification standard table, the road traffic conditions in Qingyang District of Chengdu City were judged, and the results were visualized by ArcMAP. According to the judgment results, the main findings are: (1) There are more congested sections and fewer smooth sections at morning peaks and late peaks than that in the afternoon and evening; (2) There is an obvious trend of better road traffic from the first ring to the third ring (from inside to outside) at morning peaks and late peaks. At the same time, 8 road sections (areas) that are prone to traffic congestion were identified. After analyzing theirs characteristics, it was found that: (1) It is prone to congestion in sections near traffic hotspots such as hospitals and scenic spots; (2) In terms of quantity, second-level roads are more congested. Secondary roads near the park and the residential areas are more likely to be congested.

1. Introduction

With the continuous advancement of urbanization, the prosperity of urban economy and the increase of population make the rapid development of road traffic an inevitable trend. However, the increasing demand and supply of traffic will inevitably lead to a series of urban problems caused by road traffic. In recent years, car ownership in China has been on the rise. ¹ By the end of 2018, the number of cars in China had reached 240 million, with more than 3 million in eight cities including Beijing and Chengdu. With the explosive growth of car ownership, the problem of traffic congestion has been paid more and more attention.

According to traffic analysis report of major cities in China in 2017 released by Gaode map, in 2017, more than 26 percent of Chinese cities were congested during rush hour, 55 percent were slow, and only 19 percent were unaffected by the rush hour. Meanwhile, Chengdu ranks 21st in congestion, which will not only cause additional economic loss, waste of time and energy, but also the serious environmental pollution. Hence, coordinating the relationship between road traffic development and urban environmental protection and realizing sustainable urban development has become a major problem faced by big cities in the world.

In the data age, data analysis can be used to find the causes and solutions of problems accurately, quickly and comprehensively. For the study of traffic congestion, it plays an important role to conduct data analysis using clustering algorithm. Rempe ² et al. proposed a clustering algorithm to reduce the complex traffic network to the most frequently congested part (cluster), and then analyzed the congestion behavior of the obtained clustering, which provided a basis for the potential traffic estimation and prediction system. On the basis of the traffic performance index (TPI), Wen ³ et al. proposed a dynamic macro index to reflect the congestion intensity, identified typical traffic congestion patterns by using the clustering method, and made a comparative analysis of the setting rules of different clustering indexes. Angelica ⁴ et al. proposed a method to determine intersection congestion, which used k-means algorithm to randomly select samples from images representing various degrees of traffic congestion. K-means algorithm is one of the most widely used unsupervised methods in cluster analysis, which aims to partition a set of vector inputs into K clusters through exploratory data analysis. ⁵ The above research shows that K-means is an effective method to study traffic congestion. However, in such a partition clustering algorithm, there is no uniform standard to specify the number of clusters k to be generated, and there are few related researches on Chengdu.

At the same time, with the development of a new generation of intelligent transportation system technology (such as floating vehicle traffic information acquisition technology, on-board GPS device positioning technology, etc.) ^{6, 7, 8, 9}, it becomes possible to collect and analyze large-scale traffic track data. Based on the above data, we can study the traffic congestion pattern scientifically and find out the underlying rule, which provides a basis for relevant studies such as traffic congestion prediction, and helps the traffic management department to put forward the corresponding plan, which plays a positive role in easing urban traffic congestion.

Therefore, based on real taxi GPS data (mainly speed data), this study aims to use K-means clustering method for data analysis, and establishes the traffic congestion classification standard in Chengdu urban area, based on which, the road traffic condition in Chengdu urban area is judged, and the road congestion pattern in Chengdu urban area is studied and visualized. This study aims at the actual traffic congestion in Chengdu urban area and can effectively analyze the congestion problem. Meanwhile, it provides a theoretical basis for relevant researches on road congestion detection and prediction, which would provide a new thinking for realizing sustainable urban development.

2. Data and Method

The data used in this study is the results of Guo et al. 's processing of Chengdu taxi GPS data (Urban link travel speed dataset from a megacity road network), including the vector map of Chengdu city based on OpenStreetMaps data, the "link.csv" data set and the "speed_[date]_[I].csv" data set. (Specific structure is shown in Table 1, Table 2, Figure 1).

In the research data, the Node_Start and Node_End fields in the "link.csv" data set correspond to the number of road intersection points in the vector map of Chengdu urban area. The link field in the "link.csv" dataset corresponds to the link field in the "speed_[date]_[I].csv" dataset. Therefore, the two independent data sets directly obtained above can have a certain relation based on the link field, while the vector map and the two data sets cannot be directly correlated without processing, which means that the data we get is not a comprehensive data set corresponding to the field.

The main roads, primary roads and secondary roads of the first, second and third ring roads in Wuhou district of Chengdu are studied in this paper. The selected data dates are June 8, 2015 solstice 14 (Monday to Sunday) and June 17, 2015 (Wednesday).

Clustering is the process of grouping a collection of objects into categories composed of similar objects. K-means is the most basic and simple clustering algorithm. The idea of k-means algorithm is to divide the sample set into k clusters according to the distance between samples for a given sample set. Based on the important role of clustering algorithm in data analysis and traffic congestion research, this study will use K-means for analysis.

Elbow method is one of the methods to find the best number of clusters. It regards the total sum of squares (WSS) as a function of the number of clusters, and uses the changing relationship between the total WSS and the number of clusters k to find the best number of clusters. In visual graphics, the elbow points are the optimal number of clusters.

3. Experiment Setup

To correlate the acquired data and obtain the data that can be directly analyzed in this study, the intersection points involved in "main roads, primary roads and secondary roads from the first ring road to the third ring road" in Chengdu urban area are intercepted from the vector map, and the relevant data are exported. Then, these road intersection data were matched with the "link.csv" data set to obtain the link number of the target road segment, after which, the Link number data set of the target road segment is matched with the "speed_[date]_[I].csv" data set to obtain the speed data of the target road segment with a time interval of two minutes (i.e., the two-minute speed data). Meanwhile, in order to support the subsequent research, the speed data with a time interval of two hours (i.e., the two-hour speed data) is obtained by doing arithmetic mean to the data of the corresponding time period in the "two-minute speed data".

Based on the existing research, the values of 3, 4, 5 and 6 are selected as the initial optimal clustering number k. Using Elbow method, K-means clustering calculation was performed for different k values and two-minute speed data of each target road segment every day. For each value of k, we calculate WSS in the total cluster and draw the curve of WSS, and the x-coordinate of the elbow point in the curve is the optimal clustering number k.

Then, K-means clustering is carried out. First, we selected n object data from the two-minute velocity data set of each target segment, and randomly selected k of them as the initial clustering center. Then, the similarity measure is judged based on Euclidean distance, and the cluster of each object data is determined. Next, the new cluster center for each cluster is determined by calculating and updating the average value of objects in each cluster. After calculating the criterion function E, repeat the last three steps above until the criterion function E is within the allowable error range.

First, the minimum and maximum speeds of two minutes speed data per day in each cluster categories are identified by Excel, and the arithmetic mean of the minimum speed and maximum speed in different clustering categories per day was performed to obtain the mean value. Then, the standard deviation of the minimum speed and maximum speed per day was calculated.

According to the above analysis, the average minimum and maximum speed can be used as the boundary value between each category. As long as the driving speed conforms to the road speed limit, there is no need to limit the maximum speed on the classification scale. Therefore, the standard table of traffic congestion classification is obtained.

According to the standard table of traffic congestion classification, the corresponding two-hour speed data obtained by pre-processing were used to draw the visualization map of road traffic conditions for some roads in Wuhou district through ArcMap, and then the visualization analysis was carried out. Based on the analysis results and literature, the final traffic congestion classification standard table was determined and visualized.

4. Result

The results of WSS curves are shown in Figure 2. According to the curve of WSS, the abscissa corresponding to the elbow point is mainly 4 and 5, which shows that the optimal clustering k obtained is mainly 4 or 5. In order to ensure the accuracy of the final traffic congestion classification standard Table 3, Table 4, Table 5 and Table 6 are selected as cluster numbers for K-means clustering respectively in the following study.

Table 4 are standard deviations of the daily minimum and maximum speeds in three categories, from which the following conclusions can be drawn: (1) except for the standard deviation of the maximum velocity in the third category, which is larger than 1, the others are all less than 1, which indicates that the corresponding boundary values of each category has little difference during the seven days of a week, and the classification of the velocity is basically in the same range; (2) in the third category, the road traffic condition is smooth, and the speed limit is only the road speed limit, so the speed distribution is more discrete. Therefore, the standard table of traffic congestion classification can be obtained based on this standard.

According to the results of the initial traffic congestion classification standard table, the corresponding road traffic conditions of each cluster category is determined. The initial traffic congestion classification standard table is shown in Table 5.

As shown in the Figure 3, Figure 4, Figure 5, Figure 6, the "two-hour speed data" from 8:00-10:00 and 17:00-19:00 of June 17, 2015 were selected for analysis and verification. The standard table of 3-6 traffic congestion classification was used to judge the road traffic conditions, and ArcMAP was used to visualize the results on the map. Different road traffic conditions were marked by colors as shown in the legend.

According to the visualization of road traffic conditions, it can be seen that the initial traffic congestion classification standard table is basically consistent with the actual situation. From the display effect, the indicator effect of the three levels is too simple to reasonably indicate the road traffic conditions, and the six levels of renderings are a jumble of information that cannot be immediately matched to road traffic conditions. Therefore, from the perspective of visualization, four or five levels are appropriate. In terms of the rationality of data classification, a speed rating of 5 is the best choice.

Therefore, this study selects the standard table of 5 levels of traffic congestion classification as the final standard table of traffic congestion classification, and studies the road congestion pattern in Chengdu urban area based on the table.

5. Case Study - Qingyang District Traffic Congestion Analysis

Qingyang district is located in the core region of Chengdu. This study analyzed the road traffic conditions of main roads, primary roads and secondary roads in the first, second and third ring roads of Qingyang district during the periods of 8:00-10:00, 12:00-14:00, 17:00-19:00 and 21:00-23:00 on June 17, 2015, and the road traffic patterns were studied based on the result of visualization.

Table 6 is the statistics of the number of road sections in different periods of traffic conditions in Qingyang district. In addition, ArcMAP was also used in this study to clearly and intuitively show the road congestion in Qingyang district at different periods on the map, which is shown in Figure 7, Figure 8, Figure 9, Figure 1, and the following conclusions can be obtained:

(1) Generally speaking, there are more congested roads in the morning and evening rush hours. There were 79 congested roads in the morning rush hour, 92 in the evening rush hour, and only 28 in the afternoon and evening. For slow traffic, the number of road sections in the four periods has little difference. Normal and smooth road conditions are mainly in the noon and evening. For the traffic condition of very smooth road, the number of sections in the four periods analyzed is quite small.

(2) In the morning, evening rush hours and noon period, the road traffic condition from the first ring road to the third ring road has an obvious tendency to get better. However, the difference at noon is not as significant as that in the morning and evening rush hours, which may due to better road conditions and natural conditions (such as light, etc.) at noon. As for the night, the four road conditions of congestion, slow traffic, normal traffic and smooth traffic were evenly distributed, with no significant difference.

(3) Some areas are in the congestion state during all four periods of the study (or at least three times in the congestion, the rest of the time in slow status), which are shown as purple circles in Table 7 and Figure 7, Figure 8, Figure 9, Figure 10.

As can be seen from Table 7, the areas with congestion in the four periods are mostly located in hot traffic areas such as hospitals and scenic spots, most of which are secondary roads (62.5%) and are located near parks and residential areas. This indicates that hot spots and areas with large crowds are prone to traffic congestion. Therefore, it is suggested that the managers should pay more attention to the traffic hotspots, and properly manage the secondary roads near the residential areas to improve the road traffic conditions. Meanwhile, eight regions were found to be in the congested road condition in four periods, and the analysis was made on eight regions prone to traffic congestion, which would improve the analysis of the congestion pattern, and provide a basis for traffic managers to take appropriate measures to improve traffic congestion.

6. Conclusion

This paper mainly studies the road congestion pattern in Chengdu urban area. Based on the taxi GPS data, K-means algorithm and Elbow method were used for relevant analysis, and finally the standard table of grade 5 traffic congestion classification in Chengdu was determined. Then, the road traffic condition in Qingyang district was studied, and the result was visualized on the map based on ArcMAP, so that the road traffic condition could be clearly seen. Meanwhile, we analyzes the pattern of road congestion in Chengdu urban area, and extracts 8 areas prone to traffic congestion in Qingyang district. It could put forward reasonable suggestions for traffic managers to alleviate traffic congestion, which is conducive to the sustainable development of the city.

This study made a preliminary exploration of the road congestion pattern in Chengdu urban area, and obtained some research results, but there are still some deficiencies. In the future, studies such as road congestion prediction will be carried out based on the study of road congestion pattern.

References