Real-time Traffic Models, a Great Help for Dynamic Traffic Management



Real-time Traffic Models, a Great Help for Dynamic Traffic Management

Authors

K. Friso, DAT.Mobility, J. Zantema, DAT.Mobility, L.J.J. Wismans, DAT.Mobility

Description

A real-time traffic model helps both traffic managers and travelers in the prevention of congestion on their network or route, possibly by rerouting travelers. Using a sensor network, traffic forecasts up to one hour ahead are made for an urban area.

Abstract

Reliable and accurate short-term traffic state prediction can improve the performance of real-time traffic management systems significantly. Using this short-time prediction based on current measurements delivered by advanced surveillance systems will support decision-making process on various control strategies and enhance the performance of the overall network. For example, if congestion is predicted at certain locations within the next hour, traffic measures can be taken. These may include road side measures like ramp metering, lane opening or closure, traffic light regulation, but also (personalized) in-car information like in-car navigation advice (smart routing). By taking proactive action, congestion may be prevented or it’s effects limited. Real time traffic models can be of great help in this kind of decision making.
In this paper we will present the architecture and implementation of a real-time traffic model using the macroscopic dynamic traffic assignment (DTA) model StreamLine in a rolling horizon implementation. This real-time traffic model is applied for the city of Assen (the Netherlands) where a living lab is created with sensors spread over the city. The sensors continuously gather traffic information like flows (loop detectors at traffic lights), speeds (FCD) and travel times (Bluetooth & FCD). Main elements of this real-time traffic model are estimation of an origin-destination (OD) matrix and current state estimation (i.e. starting point of short term prediction). Every 5 minutes an OD matrix is estimated based on historic OD-patterns and current traffic counts. This OD matrix is assigned dynamically to the model network using the save state file of the previous time stamp. From this rolling horizon approach (i.e. hot start) the save state file provides all information about the network initial conditions for the next simulation period. The advantage of the rolling horizon approach is that no warming-up period is needed in the dynamic traffic assignment which takes less computation time and it results in consistent results. Further, the current traffic state estimation is done by combining model estimates of previous predictions and current measurements. We tested several methods for these estimations which we will present in the paper as well as their performance. The resulting actual traffic state is compared with observed speeds and travel times for validation purposes.
This estimated OD matrix and current traffic state are then input for a short-term prediction for the next hour (in time steps of 5 minutes) resulting in predicted flows, travel times and spare capacities. As mentioned, this short-term prediction can be used for dynamic traffic management with the goal to be proactive instead of reactive. In Assen the results of the real-time traffic model are used as input for in-car navigation advice in an experimental setting.
The real-time model is also useful for travelers and can in fact be used in 3 stages: (1) pre-trip to check if it is the right time to leave, (2) on-trip for finding the optimal route and (3) post-trip for evaluation and learning if the best choice was made or what eventually a better choice could have been.

Publisher

Association for European Transport