路径选择PPT.ppt

Predictability in Traffic and Short-Term Forecasting of Traffic States Wei-Hua Lin Department of Systems and Industrial Engineering The University of Arizona Tucson, AZ 85721 weilin@sie.arizona.edu Presentation at Southwest Jiaotong University January 10, 2011  Outline Introduction Background Logic of the proposed model Model formulation Validation确定of assumptions假设 Properties and performance of the proposed model Predictability of traffic states Summary Problem Statement The problem for this study At time interval i-1 (5 to 15 minutes), given the traffic state at i-1, i-2, …, we want to predict the traffic state at time interval i. Issues to study Are traffic states predictable? If traffic states are predictable, to what extent they are predictable? Are some states more predictable than others? How far can we really go in terms of accuracy? What is the prediction accuracy we really want to pursue? Development of a method for prediction that is robust耐用的, efficient in terms of calibration标准 and computation, and reasonably accurate. Previous Study Time series model (Moorthy et al, 1988; Lee and Fambro, 1999) Kalman filter model (Okutani and Stephanedes, 1984) Non-parametric methods (Davis and Nihan, 1991, Smith and Demetsky, 1997) Simulation (Duncan and Littlejohn, 1997) Neural network approaches (Smith and Demetsky, 1994; Dochy et al, 1995; Florio, 1995, Park et al, 1998, Zhang, 2001, Abdulhai 2002) Approaches for Short-Term Forecasting Traffic With Information from Loop Detectors探测器 Two major approaches Prediction based on traffic flow theory Knowledge about the underlying基础的 traffic dynamics Propagation 传播of waves The three detector problem Requires high quality traffic data, data with finer出色的 resolution Prediction based on pattern analysis Historical and real-time point traffic information Knowledge of transition of traffic states at different time Requires less detailed traffic data The approach adopted in this study Prediction with data from