Abstract
In this study, a novel approach incorporating a two-stage prediction model with an integrated algorithm is proposed to estimate freeway dynamic origin-destination (O-D) matrices. The two-stage prediction model uses growing hierarchical self-organizing map (GHSOM) to extract clusters of traffic patterns and then uses genetic programming (GP) to predict the on-ramp traffic flow in each cluster. The integrated algorithm combines cell transmission model (CTM) with extended Kalman filtering (EKF) to estimate the arrival distributions and the O-D proportions. To demonstrate the proposed approach, a field study of on-ramp traffic patterns on a freeway stretch is examined. The results show that the proposed approach can accurately predict the traffic and satisfactorily estimate the O-D proportions along the freeway stretch.
