Efficient Evacuation Planning for Large Cities

Abstract

Given a large city represented by nodes and arcs, with a subset of these nodes having people required to be evacuated in a disaster situation, and a subset of destination nodes where the people may be taken for safety, the evacuation route planner finds routes in this network to evacuate the people in minimum possible time. Evacuation route planning in pre-known disasters such as hurricanes is of utmost importance for civic authorities. Computing such evacuation routes in a large graph with road infrastructure constraints such as road capacities and travelling times can be challenging. The Capacity Constrained Route Planner (CCRP) is a well studied heuristic algorithm proposed by Lu, George and Shekhar for solving this problem. It uses shortest path computations as a basis for finding and scheduling possible evacuation paths in the graph. However, the algorithm fails to scale to very large graphs. Other algorithms based on CCRP like CCRP++ and Incremental Data Structure based CCRP have been proposed to scale them to larger graphs. In this paper, we analyze these algorithms from performance perspective and suggest a faster algorithm by extending the CCRP algorithm that avoids recomputing partial paths. We have carried out experiments on various graph structures which show a vast improvement in runtime without impacting the optimality of CCRP results. For instance, for the city of Oldenburg with 382285 people, 6105 nodes and 7034 edges, our algorithm produced an evacuation plan in 2.6 seconds as compared to 123.8 seconds by CCRP and 9.3 seconds by CCRP++.