AETROS: Dynamic Travel Route Assignment Coping with Highly Changeable Scenarios at Traffic Networks

Dynamic Traffic Assignment is currently a key topic for the most advanced traffic management systems. Basically, it consists on automatically diverting traffic to the optimal choice of a set of alternatives – or combination of alternatives – towards each vehicle‟s destination node. A truly dynamic traffic assignment system could, potentially, cope with a highly variable traffic user demand and conditions, even with unexpected situations, like accidents. The research question for us is: Can we create an adaptive predictive architecture that provides, for each driver, the best choice at each intersection, for its destination and the current global traffic situation? In this work we share the results of an intensive set of tests accomplished with the Adaptive Evolutionary Travel Route System (AETROS). This model was designed to solve the travel route assignment problem in a new dynamic way. Urban traffic use to be highly changing scenarios where statistic based optimization approaches cannot cope with unforeseeable situations. When it is about assigning travel routes to users, it is usually approached in a static manner: when a traveler is assigned a route, it remains fixed from origin to destination. However, in the meantime the traffic scenario may suddenly change. And all those already assigned routes may became easily suboptimal. That is why AETROS is designed to be constantly adapting to the current situation and drivers are not assigned a fixed origin-destination route. Instead, as they approach an intersection they will receive the last optimal option for their destination. We have tested two different networks using a Beowulf cluster and a supercomputing facility. The smaller network was used to test whether the system correctly suggests the best option, by us introducing controlled accidents. The second bigger network has been used to deeply test the system through a wider range of traffic demands and accident probabilities. With the bigger network tests are subdivided into three different subsets. The same setup parameters where tested allowing drivers to use or not the suggested best options, forcing drivers to take the optimized options and finally without any route optimization. The presented results are meaningful, giving us important clues to keep on pushing the AETROS development.