Abstract
Contraflow lane reversal is a widely used traffic management strategy for mass evacuations, as it prioritizes outbound traffic from high-risk areas. Existing studies primarily optimize static or localized contraflow configurations, while the staged transition from normal to emergency traffic operations remains largely unexplored. In practice, infrastructure cannot be reconfigured instantaneously; operators must determine a feasible sequence of incremental modifications that gradually transforms network capacity under evolving demand. In this study, we propose a combinatorial contraflow reconfiguration approach that gradually activates contraflow links prior to the start of evacuation. To model evacuation traffic dynamics, we develop a three-dimensional, multi-region Macroscopic Fundamental Diagram (MFD) model that captures the interactions between horizontal road congestion and vertical shelter congestion. To overcome the combinatorial explosion of transition candidates, we develop a solution algorithm based on the Cross-Entropy Method with sampling-space restriction, which compactly represents admissible configurations and enables efficient sampling of feasible reconfiguration sequences using Zero-suppressed Binary Decision Diagrams (ZDDs). Computational experiments show that the ZDD-based algorithm substantially reduces calculation time relative to naive sampling. A case study in Tokyo demonstrates that optimally staged deployment reduces average travel time during and prior to evacuation by up to 18% compared with random reconfigurations, while mitigating transition-induced traffic disruption.