ChargeRL: Deep Reinforcement Learning Framework for Bidirectional Wireless EV Energy Flow Control

Abstract

The high pace of electric vehicles (EVs) adoption requires effective and smart energy management mechanisms, especially in bidirectional wireless power transfer (WPT). Conventional rule-based or optimization-based techniques cannot adjust successfully to the changing and stochastic circumstances and result in a suboptimal use of energy and quicker battery degradation. This paper demonstrates ChargeRL, a new Deep Reinforcement Learning (DRL) platform of autonomous bidirectional energy flow control in EVs. ChargeRL uses both DDPG and PPO to provide continuity in action selection that allows the agent to dynamically decide how much power to charge or discharge or discharge power depending on the battery state-of-charge (SOC) in real time, grid demand, and the availability of renewable energy. The training and evaluation of the framework were done using a simulated environment, which included stochastic EV arrivals, variations in the grid load, and battery constraints. Findings show that ChargeRL works better than traditional rule-based and heuristic strategies by making it to better optimize energy use (up to 97%), minimizes battery degradation (by 5 and 6, respectively), and increases battery life. There is also stable convergence of rewards that the framework shows, which translates to learning and generalization to unobserved situations. ChargeRL can offer a powerful, dynamic, and efficient system to next-generation smart EV charging to achieve an easy integration with renewable energy sources and grid services.