Integrating demand forecasting and deep reinforcement learning for real-time electric vehicle charging price optimization
Monowar Mahmud, Tarek Abedin, Md. Mahfuzur Rahman, Shamiul Ashraf Shoishob, Tiong Sieh Kiong, Mohammad Nur‐E‐Alam
Abstract
The rapid growth of electric vehicles (EVs) demands efficient, grid-friendly charging systems. This study introduces a dynamic pricing framework combining short-term demand forecasting and deep reinforcement learning. Using Adaptive Charging Network (ACN) data, XGBoost predicts charging demand accurately (R 2 = 0.84, MAE = 0.45 kW). Compared to a uniform rate applied to all charging usage, set at 0.15 USD/kWh across all hours, with no adjustment for system demand conditions or time-of-day, the optimized strategy enhanced total daily revenue by 133 % and diminished load variance by 72.37 %. The PPO agent also surpassed traditional Time-of-Use and demand-based pricing models by 67–94 %, while ensuring pricing stability with a price standard deviation of 0.132 USD/kWh. The simulation results illustrate the framework's efficacy in facilitating off-peak charging and improving grid reliability.