Realistic modelling of driver behaviour during evacuation scenarios is vitally important for creating effective training environments for disaster management. However, few current models have satisfactorily incorporated the level of complexity required to model the unusual driver behaviours which occur in evacuations. In particular, few state-of-the-art traffic simulators consider desires of a driver other than to travel the quickest route between two points. Whereas in real disaster settings, empirical evidence suggests other key desires such as that of being near to other vehicles. To address this shortcoming, we present an agentbased behaviour model based on the social forces model of crowds, which explicitly includes these additional factors. We demonstrate, by using a metric of route similarity, that our model is able to reproduce the real-life evacuation behaviour whereby drivers follow the routes taken by others. The model is compared to the two most commonly used route choice algorithms, that of quickest route and real-time re-routing, on three road networks: an artificial "ladder" network, and those of Louisiana, USA and Southampton, UK. When our route choice forces model is used our measure of route similarity increases by 21-169 %. Furthermore, a qualitative comparison demonstrates that the model can reproduce patterns of behaviour observed in the 2005 evacuation of the New Orleans area during Hurricane Katrina.

Accurate modelling of driver behaviour in evacuations is vitally important in creating realistic training environments for disaster management. However, few current models have satisfactorily incorporated the variety of factors that affect driver behaviour. In particular, the interdependence of driver behaviours is often seen in real-world evacuations, but is not represented in current state-of-the art traffic simulators. To address this shortcoming, we present an agent-based behaviour model based on the social forces model of crowds. Our model uses utility-based path trees to represent the forces which affect a driver's decisions. We demonstrate, by using a metric of route similarity, that our model is able to reproduce the real-life evacuation behaviour whereby drivers follow the routes taken by others. The model is compared to the two most commonly used route choice algorithms, that of quickest route and real-time re-routing, on three road networks: an artificial "ladder" network, and those of Lousiana, USA and Southampton, UK. When our route choice forces model is used our measure of route similarity increases by 21%-93%. Furthermore, a qualitative comparison demonstrates that the model can reproduce patterns of behaviour observed in the 2005 evacuation of the New Orleans area during Hurricane Katrina.

Abstract-Realistic simulation environments play a vital role in the effective training of traffic controllers to respond to large-scale events such as natural disasters or terrorist threats. BAE SYSTEMS has developed a training environment that comprises of: a physical traffic control centre environment, a 3D visualisation and a traffic behaviour model. In this paper, we describe how an agent-based approach has been essential in the development of the traffic operator training environment, especially for constructing the required behavioural models. The simulator has been applied to an evacuation scenario, for which an agent-based model has been developed which models a variety of relevant driver evacuation behaviours. These unusual behaviours have been observed occurring in real-life evacuations but to date have not been incorporated in traffic simulators. In addition, our agent-based approach includes flexibility within the simulator to respond to the variety of decisions traffic controllers can make, as well as achieving a strong degree of control for the scenario manager.

No abstract