Unreasonable Road Planning Discovered by Taxi Trajectories in Shanghai
Background: The traffic jams and congestion of the urban transportation is a problem which is not
only a social problem but also a economical one. China’s rapid urban development has resulted in
unprecedented urban population growth and built-up area expansion, but the urban traffic
development in China can not fit for the development of the urban. As the most representative
metropolitan in China, Shanghai also faces with the contradiction between the support and the
demand of urban traffic.
The push for publishing huge volumes of real-time data of taxi trajectories enhance the
effectiveness to solve the problem. Motivated by the opportunity of building a smarter city, we
came up with a vision to connect urban sensing, data management and analytics, and service
providing into a recurrent process for unobtrusive, continuous improvement on people’s life quality,
urban environment and city operation systems.
Goals: The challenge of this research lies in two aspects. The first is city-wide traffic modeling. Here,
it should be the city-wide mobility patterns of people instead of traffic conditions on a single road.
Second, we need to identify the root causes of the problem, instead of finding some observations.
We are not detecting traffic jams which are only observations. We carried out this work using the
data generated in the period of the consecutive one month, and evaluate them with Shanghai
maps and some real city planning of Shanghai.
Methods: Firstly, the whole city is partitioned into regions using major road networks which are
composed of highway and ring roads. Each region stands for a community including some
neighborhoods and low-level road segments. Traffic problems appearing on roads are just
observations while regions are the root causes of the problem, because People live in regions and
travel between regions. Further, we partition the taxi trajectories into some portion according to
traffic conditions using a data-driven method. For example, in Shanghai, work day has been into 4
time slots corresponding to morning rush hours and evening rush hours.
Then, we identify three dimensions representing the connectivity between any two regions. The
first one is the number of taxis traveling between the two regions. The other is the expectation of
the speeds of these taxis. The third one is a ratio between the average of actual travel distances
and the Euclidian distance between the centers of the two regions. We select the edges with the
traffic volume above the average, and then according to the other two features, we detect the
skyline edges aiming to find out the region pairs with a very poor connectivity, that is the edges
with a big ratio and lower expectation of the speed.
For each time slot we can obtain some skyline edges. Now, we formulate some skyline graph by
associating these skyline edges. Then each graph should be connected in chronological order
during a day. These day-long connected graphs will be broken down into chains, in order to
revealing the frequent sub-graph pattern across days. The frequent chains are counted to
understand the pattern deeply and to avoid false alert like spatially overlapping and temporally
adjacent. A chain appears more frequent means more terrible solution from the region of chain
start to the region of chain end.
Results: We detect the insufficient regions in the existing road planning of Shanghai using GPS data
of cabs travelling in the urban areas. The results are comprised of two set of findings. One is some
frequent sub-graph patterns consisting of region pairs with significant problems and the linking
among these regions. The other is the association relations between these patterns. The
distribution and trend of these three features changes over time of day on workdays and weekends
respectively. Besides, the three features (we used to detect flawed planning) well reflect on
people’s mobility patterns and traffic at a city-wide level. A lot of commonsense knowledge and
interesting stories can be found in our study.
Comparing the busy chains with policies, news and road construction in 2013, we find some
patterns leading the traffic problem. By pointing out underlying problems, the research shows
urban planners where to focus their attention.
Keyword: GPS trajectory, road planning, data mining, big data,