Approximate Event-Based Optimization for Evacuation

Evacuation is of great practical importance. It is in general hard to find the optimal evacuation guidance policy due to the uncertainty in the evacuation process, the large state and action spaces, and the lack of accurate closed-form model to describe system dynamics. In this paper, we follow our previous work to formulate the problem as an event-based optimization. We make three contributions in this paper. First, we develop action elimination technique to exponentially reduce the action space while preserving the optimal action. Second, we develop sample-path reuse technique to improve the simulation efficiency. Third, we demonstrate the performance of this approximate event based optimization algorithm on an evacuation problem. We hope this work sheds insight to improve evacuation guidance in general.