Novel and Automated Methodology to Model and Analyze Massive Evacuation along the Highway System

Unpredictable natural disasters or terrorist attacks may require that the residents of the attacked area be evacuated immediately using a partially damaged infrastructure. Evacuees may not have access to working vehicles, and may need to walk to the closest shelter or to a temporary safe place where public transit such as buses will pick them up. A portion of the population will get injured and have to go to hospitals.;An automatic process of assigning addresses to the closest hospital, shelter or safe place based on shortest distance along a potentially damaged road system has been developed. Alternatively, people with access to vehicles are assigned to exit locations on the county borders. The graph-theoretic Voronoi Diagram (GTVD) can be computed efficiently and in real time using our data structures. Guilford County data was used as an example where public schools are assumed to be the safe places. We found that the assignment was not uniform at all, and some underutilized safe places were therefore discounted, and the assignment was redone. Moreover, we create data structures that are automatically derived from Geographic Information Systems (GIS) shapefiles. The data structures emphasize the computational geometric-aspect.;A model of the highway system described in our data structure has been used to automatically develop a colored deterministic and stochastic Petri net which can then be used to simulate the dynamics of people movement on the highway system towards medical facility or exit or safe points. The basic idea behind the model is that an intersection will be a subnetwork of several queues. For example a full bidirectional 4-way stop sign intersection can be represented by nine servers whose service time is either exponential or deterministic. The intersection will serve the vehicles from the incoming traffic flow and move each vehicle to its target road segment that connects to the next intersection on the path of its destination.;We introduce a modeling procedure based on Petri nets for road intersections where each intersection consists of several places and several transitions. We introduce the concepts of places denoted as intersection-Hold (I-Hold), Fusion, Branching and road-Hold (R-Hold) places. Vehicles at an I-Hold place are inside the intersection. In a Fusion place, vehicles are waiting to enter the intersection. In the Branching place, vehicles are leaving the intersection and are assigned to their next road segments. In R-Hold place, vehilces are on the road segment. Entering the intersection is controlled by a deterministic transition that represents the arrival rate to the intersection. Leaving the intersection is controlled by a stochastic transition which is the service rate for the intersection. Similarly, entering the road segment is controlled by a deterministic transition that represents the arrival rate to the road segment and leaving the road segment is controlled by a deterministic transition which is the service rate. For example the 4-way stop sign intersection has 11 places (8 R-Hold, 1 Fusion, 1 Branching, 1 I-Hold) and 18 transitions.;We develop a mesoscopic modeling and simulation approach for modeling traffic flow over a large geographic area where people at addresses are assigned to vehicles and their destinations and routes are computed. To make the simulation specific, we consider the scenario in which most people self-evacuate GC through the closest exit points over the shortest path. Dijkstra's algorithm is used to compute the shortest paths. The simulation approach is microscopic in the sense that individual vehicles are simulated but that the highway system and the vehicle interactions are simplified.;We develop an automatic method to assemble the Petri net that represents the evacuation of Guilford County (GC). The Petri net includes 35476 places, 43540 transitions with 531595 tokens where each token represents a single person in GC. We simulate the evacuation and develop statistics and evaluation for the results. We found that the evacuation took about 8.7 hours. We locate the bottlenecks of the evacuation.