Scheduling Electrical Ambulance Charging and Patient Assignment

The need for electricity has changed the way our cities work, given both the amount of power required at any given moment but also the fluctuation of the power sources implemented these days, as the renewable power sources tend to have a fluctuation that needs support to facilitate the required outcome. The stabilization of the grid comes either from power units that are quick to react such as natural gas units or from already stored electricity. This thesis aims to enhance the efficiency and sustainability of urban emergency medical services (EMS) by integrating electric ambulances into a smart grid ecosystem, facing at the same time the need mentioned above for our modern cities, by creating with the electric ambulances a power network that is available to stabilize the power grid. Focusing on intelligent dispatch and charging scheduling, we develop a reinforcement learning (RL) framework that jointly minimizes ambulance response times and maximizes the use of renewable energy. Leveraging real-world operational data from Greece’s national EMS provider (NCEC Thessaloniki) and high-resolution EU energy production data, the system models operations at a per-minute scale with spatial-temporal preprocessing, route estimation, and green energy forecasting. A custom simulation environment built with OpenAI Gymnasium and Stable-Baselines3 trains RL agents (PPO, A2C) to make dynamic dispatch, charging, and discharging decisions via a reward function balancing speed and sustainability. Results show that optimized PPO variants significantly improve both energy management and emergency response, demonstrating the potential of AI-driven decision systems for managing electric ambulance fleets and setting the groundwork for scalable, green EMS operations.