Performance Analysis of Reinforcement Learning-Optimized Firefighting Helicopter Routing Strategies in Simultaneous Multi-Focus Wildfire Scenarios

Forest and wildland fires have been increasing in both frequency and intensity on a global scale, driven by climate change. Simultaneous multi-focus wildfire scenarios strain the capacity of conventional response planning and demand rapid critical resource allocation decisions. This study aims to optimize firefighting helicopter routing strategies using Reinforcement Learning (RL) methods. The model, developed within a Multi-Agent Deep Reinforcement Learning (MADRL) framework, was tested in simulation environments featuring simultaneous multiple fire foci. The Nearest Neighbor Heuristic (NNH), Genetic Algorithm (GA), and Mixed Integer Linear Programming (MILP) were employed as baseline comparison methods. Experimental findings revealed that the proposed MADRL-based routing strategy reduced average suppression time by 34.7% compared to NNH, 18.2% compared to GA, and 9.6% compared to MILP. Particularly in scenarios involving five or more simultaneous fire foci, the reinforcement learning model demonstrated marked superiority owing to its dynamic re-routing capability. These results highlight the potential of RL-based decision support systems in disaster management while providing direction for future studies on real-time operational integration.