Deep Deterministic Policy Gradient Data-Driven Approach for EV Charging and Discharging Scheduling Problem

This paper presents Deep Deterministic Policy Gradient approach as a model-free deep reinforcement learning method for the management of electric vehicles (EVs) charging and discharging processes through vehicle-to-grid technology in the intent to minimize the charging cost from an EV owner perspective and provide peak-shaving services to reduce the strains on the power grid infrastructure due to uncontrolled charging of EVs. The scheduling problem is formulated as a Markov Decision Process (MDP) which serves as a representation of this complex decision making problem and solved by utilizing DDPG as a data-driven approach. The goal is to learn an optimal deterministic policy that enables charging actions during valley hours and discharging during peak hours while considering the driving demand of the EV owner and the variability in driving patterns, arrival and departure times, and the EV battery state of charge.