Q-learning-Based Battery Action Decision-Making for Electric Vehicles with Target SOC

Electric vehicles (EVs) and Internet of Things (IoT) devices are becoming increasingly widespread, making efficient battery management essential. In this work, Q-learning, a reinforcement learning method, is applied to help battery-powered devices and EVs determine optimal charging, discharging, and standby strategies. Through continuous learning, the battery explores different actions and improves its decision-making process over time. The results demonstrate significant improvements: the average state value increased from 90.83 to 98.70 (+8.66%), while the standard deviation decreased from 18.96 to 6.39 ($66.27 \%$), indicating a more stable and efficient decision-making process. This reduction in variability highlights the method’s ability to enhance charging decisions, reducing fluctuations and improving overall battery performance. The proposed approach contributes to more effective battery management, potentially supporting advanced energy systems such as vehicle-to-grid (V2G) integration.