A new model was proposed to estimate dynamic origin-destination (O-D) travel time and flow simultaneously on freeway corridors using a Neural-Kalman filter (NKF). The model yields O-D travel time and flow in real time from link traffic counts, spot speeds and off-ramp volumes measured at observation points along the freeway. State and measurement equations, which play important roles in the Kalman Filter (KF), were newly defined by artificial neural network (ANN) models without assuming any analytical functions. An extended KF model was modified to take the influence of state variables into account for some previous time steps. Also, a macroscopic traffic flow model was integrated with the NKF model to predict traffic states on the freeway corridors in advance. Numerical analyses showed that the use of the NKF model as well as the macroscopic model was effective in estimating dynamic O-D travel time and flow more accurately. (A*)
Abstract