Dynamic Incentive Pricing Menu Differentiated EV Real-Time Charging Schedule to Increase Renewable Energy Consumption: A Deep Reinforcement Learning Approach

Unregulated charging for a large fleet of electric vehicles places excessive strain on the power grid during peak periods. The successful implementation of an aggregator‐based electric vehicles (EVs) demand response program in the smart grid relies on the incentive pricing scheme and the charging schedule protocols. It is challenging for the aggregator to make real-time decisions due to the existence of randomness in traffic conditions, EV driver flexibility, and wholesale electricity markets. In this study, the real-time EV charging rescheduling problem is formulated as a modified Markov decision process-based reinforcement learning approach to determine the dynamic price menu offered to the individual EV driver and the paired optimal charging arrangements. The case study demonstrates that the dynamic price bidding structure compensates the drivers for the inconvenience of having the EV plugged in for an additional amount of time. The dynamic charging rescheduling program creates remarkable financial benefits for the aggregator and the EV owners. The modified Q-learning policy with an adaptive data-driven discretization technique can address the issue of the computational burden caused by huge action spaces in reinforcement learning applications.