Abstract
Over the past decades, the increase in both frequency and intensity oflarge-scale wildfires due to climate change has emerged as a significantnatural threat. The pressing need to design resilient landscapes capable ofwithstanding such disasters has become paramount, requiring the development ofadvanced decision-support tools. Existing methodologies, including MixedInteger Programming, Stochastic Optimization, and Network Theory, have proveneffective but are hindered by computational demands, limiting theirapplicability. In response to this challenge, we propose using artificial intelligencetechniques, specifically Deep Reinforcement Learning, to address the complexproblem of firebreak placement in the landscape. We employ value-function basedapproaches like Deep Q-Learning, Double Deep Q-Learning, and Dueling DoubleDeep Q-Learning. Utilizing the Cell2Fire fire spread simulator combined withConvolutional Neural Networks, we have successfully implemented a computationalagent capable of learning firebreak locations within a forest environment,achieving good results. Furthermore, we incorporate a pre-training loop, initially teaching our agentto mimic a heuristic-based algorithm and observe that it consistently exceedsthe performance of these solutions. Our findings underscore the immensepotential of Deep Reinforcement Learning for operational research challenges,especially in fire prevention. Our approach demonstrates convergence withhighly favorable results in problem instances as large as 40 x 40 cells,marking a significant milestone in applying Reinforcement Learning to thiscritical issue. To the best of our knowledge, this study represents a pioneering effort inusing Reinforcement Learning to address the aforementioned problem, offeringpromising perspectives in fire prevention and landscape management