COMPARISON OF PREDICTION METHODS FOR URBAN NETWORK LINK TRAVEL TIMES

摘要

Traffic congestion is becoming a serious environmental threat that must be resolved quickly. The mobile travel information system developed at The Nottingham Trent University enables the integration of data concerning traffic flows and individual journey plans thus making it possible to perform optimisation of travel. This paper focuses on the issue of provision of real-time information about urban travel and assistance with planning travel. Nottingham's SCOOT (Split Cycle Offset Optimisation Technique) traffic-light control system provides real-time information about the link travel times within certain areas of the city. However, rather than using link travel times at the time of the request, it is more effective to predict the link travel times for the time of travel along the particular links. The future link travel times depend upon the historical travel time of the link (for the specific time step in the day) as well as the current link travel time. Consequently, the link weights are a combination of real-time data, historical data and static data. Three prediction methods have been implemented and tested in the context of Nottingham's urban road network. The preliminary results suggest that the information discounting technique gives the best results.