01516077

Component

Dynamic route planning is a promising way to improve the quality of route planners in navigation systems based on real-time traffic data. Real-time traffic data offer information for predicting fluctuations and changes in traffic speeds on links of a road network which routing algorithms can take into account. The goal of this study is to introduce and test a system of incremental learning of link speed profiles based on real-time traffic data. In the proposed system, day-by-day fluctuation in traffic conditions on a link is represented by a discrete distribution of speed profiles. Learning the distribution is formulated as a dynamic version of the K-means clustering problem. Based on a dynamic K-means algorithm the system learns the discrete distribution (K means) incrementally and continuously. Furthermore, in the short term, the system continuously updates its current belief of the actual realization of the link’s speed profile based on continuous speed measurements from a sensor. The system is tested based on numerical experiments where sensor data are simulated. The results indicate that, in the long-term, the system effectively learns assumed distributions of speed profiles and, in the short term, adequately responds to real-time traffic information to improve short-term traffic speed prediction. Therefore, it is concluded that the system offers a promising way to implement dynamic routing in current route planners for navigation systems.

This paper was sponsored by TRB committee ADB30(6) Paper Review Group #2.

01503729

14-1226

English

17p

2014

Transportation Research Board 93rd Annual Meeting

Digital/other

Figures; References

Algorithms; Automobile navigation systems; Real time information; Routing; Traffic data; Traffic forecasting; Traffic speed

Route planning

Data and Information Technology; Highways; Planning and Forecasting; I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2014 Paper #14-1226

TRIS, TRB, ATRI

Jan 27 2014 2:28PM