ENHANCING TRAVEL TIME FORECASTING WITH TRAFFIC CONDITION DETECTION

摘要

Short-term traffic forecasting aims to provide more reliable travel information service, so as to assist people in making more reasonable travel decisions. With the increasing availability of traffic data along with the development of communication technology, both the capability and accuracy of travel time forecasting have been significantly enhanced in real-time conditions and a great number of forecasting methods have been carried out during recent years. However, they are inadequate when confronted with the real world traffic problems, since the real-time traffic condition can be affected easily and changed constantly. In our study, a hybrid forecasting approach is presented from a more practical perspective, based on a traffic condition detection method which monitors the real-time traffic condition and performs the travel time forecasting according to different traffic conditions. In particular, we first build a traffic conditions evaluation system to detect different sorts of traffic conditions. In this study, the traffic conditions are divided into four types including light, stable, congested and abnormal traffic condition according to travel time cost. We use a clustering tool to obtain traffic flow patterns of different traffic conditions. And the process characterize the state of the system with respect to the deviation of current conditions from an expected ones based on historical data as a definition for abnormalities in the traffic stream. Then the hybrid forecasting approach, in which several methods are used to deal with different traffic conditions, is trained to judge with certain confidence which method performs the best according to the certain traffic condition with historical traffic data. Then the travel time forecasting is taken out after the detection of real-time condition by the hybrid forecasting approach with fixed historical data and received real-time traffic information. Case studies are carried out using a real-time traffic dataset in downtown Beijing.