Emergency vehicle rescue route optimization based on deep reinforcement learning in highway network

Rescue route planning for emergency vehicles (EVs) is one of the most important tasks for emergency response. However, traditional methods face challenges in addressing the complex information in modern traffic environments. To ensure the prompt arrival of EVs, this paper proposes a deep reinforcement learning-based method to optimize the rescue routes of EVs on the highway. Specifically, to align with the emergency transportation business, an emergency transportation map based on the road network node graph is first constructed. To improve the convergence speed of the algorithm, the action masking technique is applied to reduce the dimension of the action space during the decision-making of the agent to avoid invalid exploration. Additionally, to help the agent learn the spatio-temporal benefits throughout the entire dispatch cycle, road network nodes are prioritized based on the spatial transfer scale, and a corresponding reward scheme is designed. These improvements, when combined with the agent's learning mechanism, can assist them optimize rescue routes in a more efficient action space. Validation is carried out through simulation examples, and the results show that the approach presented in this paper can find the shortest time-consuming rescue route. Furthermore, the proposed improvements enhance the convergence performance of the algorithm.