Agent-based models help identify path of least resistance in hurricane evacuations

July 10, 2012

Emergency managers would do well to consider staggering evacuations in a disaster like a hurricane, perhaps advising people living nearest the storm surge and neighborhoods with a lot of elderly residents to leave first, followed by advisories to less exposed people, with the idea of maintaining a steady traffic flow on the roads out.

The alternative, as experienced in the evacuation of Houston when hurricane Rita struck Texas in 2005, is evacuation routes that become more parking lot than highway.

"If everybody evacuates at the same time, no one evacuates," says Purdue Professor Satish Ukkusuri. "But if we can understand how households react to evacuation warnings and make evacuation decisions, we can build better evacuation strategies."

With assistance from Purdue's community clusters, the civil engineering professor is developing a battery of advice for emergency managers. He’s also developing techniques for predicting traffic flow at specific locations in real time, using very large-scale geo-location data, and for automatically controlling traffic signals to optimize vehicle movement.

Ukkusuri, who leads the Interdisciplinary Transportation Modeling and Analytics Lab at Purdue, also is examining such issues as where to locate charging stations for electric vehicles as their popularity grows and how to better identify and mitigate vehicle emissions hotspots in urban areas.

His lab uses Repast, a Java-based modeling and simulation system designed to work on clusters. The models are agent based, which in Ukkusuri’s case means he’s simulating the actions of individual households and drivers in a scenario.

That requires efficient algorithms, a lot of fast processing power and ample storage for the results on which Ukkusuri builds his recommendations. For example, in a hurricane simulation for Miami and surrounding Dade County in Florida there are four million households to consider and, if a quarter of them leave to flee a storm, a million vehicles on the road, each acting independently according to a set of rules built into the model.

The hurricane project is funded by the National Science Foundation and has a goal of developing real-time and planning models public policymakers can employ to improve how evacuations are carried out.

The simulations incorporate more than typical engineering transportation models. They also include social science survey data from residents in regions impacted by hurricanes Ivan in 2004 and Katrina and Rita in 2005. Ukkusuri is working with researchers at Florida International University and Virginia Tech. The idea is to factor in the behavioral framework people use in making evacuation decisions. Among the things the results reveal so far: non-homeowners are more likely to evacuate, as are people with children.

The researchers also are examining factors that impact the communication of evacuation information and how different segments of a population may rate it as being trustworthy or not depending on where they get it, whether the media in some cases or, say, a pastor or friends in others.

"We want to understand how that information spreads within the social network," Ukkusuri says. "It makes a very big difference who they get this information from."

He sees promise in combining other large "people" data sets — including mobile phone location data and location data posted to social networking sites like Facebook and Twitter — with traditional data like traffic counts to do even more sophisticated traffic network modeling. The uses could be myriad, from figuring out which restaurants aren’t too busy to get a seat tonight to specifying the number of taxis to dispatch to a locale.

Ukkusuri says the community clusters offer him two major advantages. Since capacity is shared among the faculty partners when not in use, he can typically tap additional computing power when his demanding models need it. He's also relieved of the burden of operating his own cluster, which he did in the past.

"I did not want to have someone in my lab manage the system on a day-to-day basis," Ukkusuri says. "I know how much time it takes and I don't want to do it again."

More Information

  • Hasan, S. and Ukkusuri, S.V. (2011). "A contagion model for understanding the propagation of hurricane warning information." Transportation Research Part B (Methodological), 45(10), pp. 1590-1605.
  • Hasan, S., Ukkusuri, S.V., Gladwin, H. and Murray-Tuite, P. (2011). "A behavioral model to understand household level hurricane evacuation decision making." ASCE Journal of Transportation Engineering, 137(5), pp. 341-349.
  • Murray-Tuite, P., Yin, W., Ukkusuri, S.V. and Gladwin, H. "Changes in evacuation decisions between hurricane Ivan and Katrina." Accepted for Publication in Journal of Transportation Research Record.

Originally posted: July 1, 2014  4:13pm