Vulcano: Operational Fire Suppression Management Using Deep Reinforcement Learning

Vulcano is a fire-management system based on deep reinforcement learning (DRL). Using simulated trajectories from a state-of-the-art simulator, agents are trained to select areas that should be treated to minimize fire propagation. We focus on the operational problem where fire suppression teams are deployed after detecting an ignition and collaborative strategies are critical to contain the fire. We propose a new algorithm based on centralized training with decentralized execution, modifying the reward and advantage functions to provide each agent with critical information about the teams. Experiments demonstrate the performance of the method compared to traditional approaches.