Fire Truck Firefighting Path Planning Based on Prediction and Multi-Agent Reinforcement Learning

With the increasing global warming and extreme weather conditions, fires are becoming more frequent worldwide. Deep learning can be applied to fire prediction, while reinforcement learning can be applied to firefighting truck path planning. This paper proposes a multi-agent reinforcement learning (MARL) framework incorporating a prediction module for planning firefighting truck routes. To achieve this goal, we utilize Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to forecast future fire conditions from a temporal perspective, followed by employing MARL to plan firefighting truck paths based on both the current and future fire states. Furthermore, from a spatial perspective, we integrate MARL algorithms incorporating opponent modeling into firefighting path planning, enabling firefighting trucks to make action selections by considering predictions of other firefighting trucks' actions. Using the Mao-Xianmin model to simulate fire spread environment, our results demonstrate that our approach achieves effective firefighting path planning.