Pattern-based short-term traffic forecasting for urban heterogeneous conditions

Short-term traffic flow forecasting plays a significant role in the Intelligent Transportation Systems (ITS), especially for traffic signal control and transportation planning research. Two mainly problems restrict the forecasting of urban freeway traffic parameters. One is the freeway traffic changes non-regularly under the heterogeneous traffic conditions, and the other is the successful predictability decreases sharply in multiple-steps-ahead prediction. In this paper, we present a novel pattern-based short-term traffic forecasting approach based on the integration of multi-phase traffic flow theory and time series analysis methods. For the purpose of prediction, the historical traffic data are classified by the dynamic flow-density relation into three traffic patterns (free flow, synchronized and congested pattern), and then different predict models are built respectively according to the classified traffic patterns. With the current traffic data, the future traffic state can be online predicted by means of pattern matching to identify traffic patterns. Finally, a comparative study in a section of the Third-Loop Freeway, LIULIQIAO, Beijing city, shows that the proposed approach represents more accurately the anticipated traffic flow when compared to the classical time series models that without integration with the traffic flow theory.