Adaptive opening times for evacuation shelters during disasters

Hanwen Liu,Qi Luo,Yongjia Song

DOI: https://doi.org/10.1007/s11590-024-02168-z

IF: 1.5288

2024-11-27

Optimization Letters

Abstract:The increasing frequency and severity of extreme weather events, such as hurricanes and tropical cyclones, address the importance of prompt and effective natural disaster response strategies. Our research aims at developing a learning-based decision-making framework tailored for evacuation shelter opening time (ESOT), with a focus on prioritizing the demands of vulnerable populations. This approach seamlessly integrates various complex supply- and demand-related factors, including evacuation demand forecasting and shelter operations requirements. The shelter opening time is formulated as a multi-class optimal stopping problem, which readily addresses the trade-off between the risks of false alarms and the perilous consequences of delayed responses, accommodating the uncertainties in disaster state evolution. To improve the computational and sample efficiency, we created a hierarchical policy approximation approach, providing provable optimality guarantees. Through a case study of Hurricane Florence in 2018 using historical wind speed data, our findings demonstrate the efficiency and flexibility of the ESOT policy, clearly outperforming standard stochastic optimization methods. For example, the total cost saving using our approach ranges from 6.6 to 28.2%, and the cost saving is more significant when the variance of the predictor is larger. These results highlight the benefits of integrating learning-based disaster management strategies with physics-informed forecasting models for protecting vulnerable populations in the face of disasters.

mathematics, applied,operations research & management science