Estimation of time-varying origin-destination flows from traffic counts: A neural network approach

A dynamic model based on the error back-propagation learning principle in neural network theory is proposed for estimating origin-destination flows from the road entering and exiting counts in a transportation network. The origin-destination flows in each short time interval are estimated through minimization of the squared errors between the predicted and observed exiting counts which are normalized using a logistic function. Two numerical experiments are conducted to evaluate the performance of the proposed model; one uses a typical four-way intersection, and the other one uses a real freeway section. Numerical results show that the back-propagation based model is capable of tracking the time variations of the origin-destination flows with a high stability.