Towards DQN Reinforcement Learning for energy management for bidirectional charging of EV’s

Abstract

This study applies Deep Q-Network (DQN) reinforcement learning to optimize bidirectional EV charging in a microgrid with dynamic pricing and renewable energy. The environment includes an EV, wind turbine, stationary battery, flexible household loads, and grid connection. DQN agents learn to minimize energy costs by charging during low-price periods and discharging during high-price windows. Simulations across four scenarios show improved cumulative rewards and grid efficiency. Future work will address stochastic elements, realistic EV availability, and continuous action spaces to enhance adaptability and performance in real-world applications.