Closed-Loop Planning for Disaster Evacuation with Stochastic Arrivals

Effective evacuation efforts can save lives during natural disasters. Uncertainty makes planning optimal evacuation routes difficult. Most current approaches use open-loop deterministic linear programming and integer programming. Robust programming variants have also been proposed. In this paper, we frame the evacuation route planning problem as a Markov decision process (MDP). We solve the MDP approximately using deterministic mixed-integer programs (MIPs) solved in a closed-loop fashion. We benchmark this policy against the optimal MDP policy where tractable. We also solve deterministic integer programs in an open-loop fashion to compare against our closed-loop MIP solutions. Closed-loop integer programming techniques are shown to obtain up to 90% of the performance of the optimal MDP policy, and can outperform open-loop approaches by as much as 52%. Performance is measured in terms of number of lives saved.