Seminars - ABSTRACT
Shashi Shekhar, Ph.D., McKnight Distinguished University Professor
Department of Computer Science, University of Minnesota
Title: Evacuation Route Planning: Novel Spatio-temporal Network Models and Algorithm
Abstract: Efficient tools are needed to identify routes and schedules to evacuate affected populations to safety in face of natural disasters or terrorist attacks. Challenges arise due to violation of key assumptions (e.g. stationary ranking of alternative routes, Wardrop equilibrium) behind popular shortest path algorithms (e.g. Dijktra's, A*) and microscopic traffic simulators (e.g. DYNASMART). Time-expanded graphs (TEG) based mathematical programming paradigm does not scale up to large urban scenarios due to excessive duplication of transportation network across time-points. We present a new approach, namely Capacity Constrained Route Planner (CCRP), advancing ideas such as Time-Aggregated Graph (TAG) and an ATST function to provide earliest-Arrival-Time given any Start-Time. Laboratory experiments and field use in Twincities for DHS scenarios (e.g. Nuclear power plant, terrorism) show that CCRP is much faster than the state of the art. A key Transportation Science insight suggests that walking the first mile, when appropriate, may speed-up evacuation by a factor of 2 to 3 for many scenarios. Geographic Information Science (e.g. Time Geography) contributions include a novel representation (e.g. TAG) for spatio-temporal networks. Computer Science contributions include graph theory limitations (e.g. non-stationary ranking of routes, non-FIFO behavior) and scalable algorithms for traditional routing problems in time-varying networks, as well as new problems such as identifying the best start-time (for a given arrival-time deadline) to minimize travel-time.
Experiments with real and synthetic transportation networks show that the proposed approach scales up to much larger networks, where software based on linear programming method crashes. Evaluation of our methods for evacuation planning for a disaster at the Monticello nuclear power plant near Minneapolis/St. Paul Twin Cities metropolitan area shows that the new methods lowered evacuation time relative to existing plans by identifying and removing bottlenecks, by providing higher capacities near the destination and by choosing shorter routes. In 2005, CCRP was used for evacuation planning (transportation component) for the Minneapolis-St. Paul twin-cities metropolitan area. It facilitated explorations of scenarios (e.g. alternative locations and times) as well as options (e.g. alternative transportation modes of pedestrian and vehicle). It also led to an interesting discovery that walking able-bodied evacuees (instead of letting them drive) reduces evacuation time significantly for small area (e.g. 1-mile radius) evacuations.
In future work, we plan to formally characterize the quality of solutions identified by the CCRP approach.
We will explore new ideas, e.g. phased evacuations and contra-flow, to further reduce evacuation times. In addition, we would like to improve modeling of other transportation modes such as public transportation.
[Back to Seminars Page]