An Agent-Based Model for Post-Disaster Resource Allocation Considering Dynamic Demand

In the response phase of a disaster, effective and equitable resource allocation is critical for meeting the needs of affected populations and maintaining the social order in disaster-stricken areas. However, dynamic demand driven by the mobility of disaster-affected populations can substantially impact resource allocation outcomes. To consider this mobility in the resource allocation problem, in this paper, we employ an agent-based model to simulate cross-shelter movements and assess the performance of resource allocation strategies considering a gradual increase of total available resources over time. A novel resource allocation strategy is proposed, which leverages demand estimation through a machine learning (ML) method-XGBoost. We then compare this approach against two conventional strategies: uniform distribution and historical data averaging. Through case studies using real-world data of Typhoon Ragasa from Shenzhen in China and Lal-lo in the Philippines, we demonstrate that our proposed strategy comprehensively outperforms the other two in terms of effectiveness, equity and stability. In conclusion, this ML-based prediction approach offers valuable insights to support decision-makers in the formulation of resource allocation plans.