Energy Management in Hybrid Electric Vehicles: A Q-Learning Solution for Enhanced Drivability and Energy Efficiency

Abstract

This study presents a reinforcement-learning-based approach for energy management in hybrid electric vehicles (HEVs). Traditional energy management methods often fall short in simultaneously optimizing fuel economy, passenger comfort, and engine efficiency under diverse driving conditions. To address this, we employed a Q-learning-based algorithm to optimize the activation and torque variation of the internal combustion engine (ICE). In addition, the algorithm underwent a rigorous parameter optimization process, ensuring its robustness and efficiency in varying driving scenarios. Following this, we proposed a comparative analysis of the algorithm’s performance against a traditional offline control strategy, namely dynamic programming. The results in the testing phase performed over ARTEMIS driving cycles demonstrate that our approach not only maintains effective charge-sustaining operations but achieves an average 5% increase in fuel economy compared to the benchmark algorithm. Moreover, our method effectively manages ICE activations, maintaining them at less than two per minute.

Keywords

• hybrid electric vehicles (HEVs)
• drivability
• fuel economy
• energy management
• reinforcement learning (RL)