Intelligent Charging and Delivery Management in EV Fleets Through Reinforcement Learning

In the context of the global transformation towards sustainable mobility, the management of electric vehicle (EV) fleets poses new challenges related to limited battery capacity, dynamic demands, and limited charging resources. This study proposes an intelligent approach for management optimization using the Q-learning technique. A simulation environment is created in which the agent makes real-time decisions based on the current state of the EV, charging, station load, and request priority. The approach includes options for actions such as delivery, waiting, request skipping, and adaptive charging. The results of multiple iterations show that the agent successfully minimizes the number of missed requests, maintains high overall utility, and ensures balanced resource utilization. Analysis of logs, Q-values, and visualizations confirms the effectiveness of the reinforcement learning and stabilization strategy. The study demonstrates the potential of learning to address complex logistical problems and offers a feasible framework for implementation in real EV fleets.