Journal article
Adaptive opening times for evacuation shelters during disasters
Optimization letters, Vol.19(7), pp.1399-1420
09/2025
DOI: 10.1007/s11590-024-02168-z
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.
Details
- Title: Subtitle
- Adaptive opening times for evacuation shelters during disasters
- Creators
- Hanwen Liu - Clemson UniversityQi Luo - University of IowaYongjia Song - Clemson University
- Resource Type
- Journal article
- Publication Details
- Optimization letters, Vol.19(7), pp.1399-1420
- DOI
- 10.1007/s11590-024-02168-z
- ISSN
- 1862-4472
- eISSN
- 1862-4480
- Publisher
- SPRINGER HEIDELBERG
- Grant note
- Division of Civil, Mechanical and Manufacturing Innovation: CMMI-2308750, CMMI-2045744 National Science Foundation
This work was supported by the National Science Foundation: awards CMMI-2308750 and CMMI-2045744.
- Language
- English
- Electronic publication date
- 11/18/2024
- Date published
- 09/2025
- Academic Unit
- Business Analytics
- Record Identifier
- 9984749759502771
Metrics
55 Record Views