Autonomous Scheduling of Electric Vehicle Charging in Renewable Microgrids Using Deep Reinforcement Learning

Abstract

This study outlines an autonomous charging schedule for electric vehicles (EVs) within a renewable microgrid that uses deep policy learning. Renewable energy-based micro-grid energy management systems are combined with a reinforcement learning agent that balances demand for EV charging, PV production and battery storage (i.e., variable energy sources) through EV charging identified by an agent observing the state of the microgrid, renewable resource forecasts and vehicle arrival profiles in order to select charge rates that result in minimum cost, peak demand reduction and maximization of renewable resource use. Simulation experiments of realistic load and generation time histories indicate that the proposed strategy produces a better utilization of renewable energy than heuristic scheduling methods, as well as providing potential cost savings from improving renewable energy utilization. The simulation results also demonstrate that the scheduling strategy is capable of making continuous adaptions to the variability in loads, generation and constraints of the microgrid while still meeting the charging demands of users and maintaining compliance with microgrid regulations. Through its scalability, capability to accommodate multiple vehicles simultaneously and ability to adapt continuously to changes in operating conditions, this framework provides lower operating costs, reduced emissions and increased resilience for renewable energy-based micro-grids. Future research will investigate methods for coordinating multiple agents and deploying the framework in the field. Extensive ablation studies will help quantify the contribution of each component and determine the stability of the charging policy under various scenarios.