DeepTraffic develops a new generation of traffic prediction methods to help traffic professionals make better decisions. By combining traffic flow theory, artificial intelligence and real-time data, we aim to make predictions more accurate, robust and explainable.

Road traffic is a complex dynamic system. A relatively small disruption — an accident, roadworks, extreme weather or a change in traffic demand — can trigger congestion that spreads rapidly through a network.

Reliable traffic predictions could help traffic managers anticipate these developments and intervene before local problems become network-wide disruptions. Yet predicting traffic is difficult. Traffic dynamics are nonlinear, uncertain and highly dependent on changing conditions and human behaviour.

New Generation

To make traffic predictions better, DeepTraffic investigates and develops a new generation of traffic prediction methods that combine two approaches that are traditionally used separately: a physics-oriented approach, based on traffic flow theory, and a data-driven approach, using artificial intelligence.

Physics-based models provide structure, domain knowledge and physical consistency. Machine learning can learn complex patterns and behaviours from data. DeepTraffic brings these strengths together, while explicitly accounting for uncertainty and making the resulting predictions understandable to the people who use them.

The ultimate goal is not to create a ‘crystal ball’ for traffic. It is to provide better-informed predictions and decision support, particularly when traffic conditions are uncertain, unusual or disrupted.