Urban evacuations are crucial life-saving procedures that take place in response to the thread or occurence of major disasters, aiming safely relocate large population groups to safe areas. Evacuation planning involves the consideration of financial, operational and managerial aspects of the process, and necessitates an understanding of human behavior.
When executing an evacuation, a key concern that arises is the management of the emerging, unique traffic patterns that arise due to sudden surge in traffic flows, which would usually contravene nominal traveller behaviours. These delays may be further aggravated by physical deterioration of the underlying infrastructure, caused by the event that triggered the disruption in the first place.
To solve this problem, TSL has been developing a family of evacuation routing algorithms that optimise evacuee travel through the transport network using a combined system of demand management and signal phasing. This research is led outcomes by TSL researcher Jose Escribano and is featured in an upcoming publication.
The proposed framework is structured as a Stackelberg game acting as a simulation-optimisation framework. A dynamic network loading model based on a continuous time link-based kinematic wave model evaluates the effects of flow propagation and congestion, including physical queuing and vehicle spillback.
We tested our technique on the family of Sioux Falls network instances (a commonly used benchmark dataset), with particular consideration of scalability and accuracy. Outcomes of our analysis indicate that our algorithms outperformed other commercial and off-the-shelf algorithms, providing safer, faster and more orderly evacuations.
During our study we further modelled the progress of evacuation under different network conditions, involving a combination of shelter configurations, network damage scenarios and use of strategic signal control. Results show that our algorithm consistently provides more efficient solution than the baseline, allowing more vehicles to evacuate in time.