Learning-Assisted Energy Optimization of Electric Vehicle Fleets in Renewable-Rich Microgrids

Abstract

In this paper, we propose a learning-enabled framework to support the energy management of EV fleets operating in renewable-rich microgrids. The framework leverages a combination of Reinforcement Learning, Model Predictive Control, and Probabilistic Renewable Forecasting to optimize the charging schedules of EVs and the V2G interactions between EVs and the grid to achieve minimum operating costs, reduce peak demand, and maximise the use of renewable energy. The framework has a hierarchical architecture that separates the fleet-level and the charger-level decisions, allowing for scalability and robustness against forecast uncertainty. An online learning agent utilises a reward signal to balance the price of electricity, the degradation of the battery and the constraints imposed by the grid. Comprehensive simulation results on a representative microgrid with significant photovoltaic generation show that the cost of energy and peak load have been reduced, while the utilisation of renewable energy has increased. Sensitivity analyses quantify the effect of forecast accuracy, fleet size, and market signals. The simulations demonstrate that integrating learning-based decision making into microgrid controllers provides a viable solution for efficient EV fleet management and renewable energy integration without compromising grid stability and scalability.