Reliable real-time framework for short-term freeway travel time prediction

It is widely acknowledged that traffic information has the potential of increasing the reliability in road networks and in alleviating congestion and its negative environmental and societal side effects. However, for these beneficial collective effects to occur, reliable and accurate traffic information is a prerequisite. Building on previous research, this article presents a reliable framework for online travel time prediction for freeways, which could, for example, be used to generate traffic information messages on so-called dynamic route information panels on freeways. Central in this framework is a so-called state-space neural network (SSNN) model, which learns to predict travel times directly from data obtained from real time traffic data collection systems. In this article we show that by using an ensemble of SSNN models also a measure for the reliability of each prediction can be produced. This enables traffic managers to monitor in real time the reliability of this system without actually measuring travel times.