A Q-Learning-Based Optimization Framework for Cost-Efficient and Grid-Friendly Energy Management of Electric Vehicle Fleets

The increasing adoption of Shared Autonomous Electric Vehicles (SAEVs) presents new challenges in energy management, particularly in optimizing charging and discharging schedules to minimize costs while maintaining grid stability. This paper proposes a novel Q-learning-based optimization framework designed to efficiently manage the energy consumption of SAEV fleets. The framework integrates dynamic electricity pricing and vehicle-to-grid (V2G) discharge strategies, enabling real-time decision-making that reduces operational costs by up to 43% in high electricity price fluctuation scenarios. Compared to traditional methods such as Mixed Integer Linear Programming (MILP), which typically achieve around 30% cost reduction but face scalability issues, our model demonstrates superior adaptability and efficiency in dynamic, large-scale environments. The framework operates within a comprehensive set of operational constraints, including battery health, grid load management, and safety. Numerical simulations demonstrate the model's effectiveness in minimizing energy costs and preventing grid overload, while sensitivity analysis explores the impact of key parameters such as battery capacity and charging power limits. The proposed framework offers a scalable, adaptive solution for managing SAEV energy use in increasingly complex and dynamic electricity markets, contributing to more sustainable and cost-effective fleet operations.