Identifying Urban Traffic Congestion Pattern from Historical Floating Car Data

Abstract

With the increasing amount of traffic information collected through floating car data, it is highly desirable to find meaningful traffic patterns such as congestion patterns from the accumulated massive historical dataset. It is however challenging due to the huge size of the dataset and the complexity and dynamics of traffic phenomena. A novel floating car data analysis method based on data cube for congestion pattern exploration is proposed in this paper. This method is different from traditional methods that depend only on numerical statistics of traffic data. The view of the event or spatial-temporal progress is adapted to model and measure traffic congestions. According to a multi-dimensional analysis framework, the traffic congestion event is first identified based on spatial-temporal related relationship of slow-speed road segment. Then, it is aggregated by a cluster style to get the traffic pattern on a different level of detail of spatial-temporal dimension. Aggregated location, time period and duration time for recurrent and important congestions are used to represent the congestion pattern. We evaluate our methods using a historical traffic dataset collected from about 12000 taxi-based floating cars for one week in a large urban area. Results show that the method can effectively identify and summarize the congestion pattern with efficient computation and reduced storage cost.