Unified Deep Reinforcement Learning Energy Management Strategy for Hybrid Electric Vehicles

A team of researchers from the Beijing Institute of Technology has proposed a novel energy management strategy for hybrid electric vehicles, leveraging deep reinforcement learning and meta-learning techniques to adapt to diverse vehicle types and powertrain configurations. The strategy, inspired by online hard sample mining, aims to enhance training efficiency and achieve satisfactory performance with minimal sample training. The research demonstrates a significant improvement in convergence efficiency, with a 40% enhancement, while achieving comparable final performance metrics.

Key Takeaways:

• The proposed energy management strategy employs meta-reinforcement learning to simultaneously learn energy management systems for multiple vehicle types across various operating scenarios.
• The strategy utilizes hard sample mining to optimize the presentation of random operating scenarios during training, enhancing training efficiency through a scientifically informed approach.
• The research demonstrates a significant improvement in convergence efficiency, with a 40% enhancement, while achieving comparable final performance metrics.
• The proposed strategy is validated on a simulated vehicle emulator, showcasing its effectiveness in diverse powertrain configurations and operating scenarios.
• The research team, led by Xiaokai Chen, comprises Zhiming Wu, Qianhui Li, Hamid Reza Karimi, and Zhengyu Li.
• The study highlights the potential of meta-learning and online hard sample mining in developing adaptable energy management systems for hybrid electric vehicles.

Statistics:

• The proposed energy management strategy demonstrates a 40% improvement in convergence efficiency.
• The research is validated on a simulated vehicle emulator, showcasing its effectiveness in diverse powertrain configurations and operating scenarios.
• The study employs a meta-reinforcement learning approach, leveraging both meta-learning and online hard sample mining techniques.
• The research team published their findings in the Journal of Engineering, October 13, 2025, as part of the Control Engineering Practice.

Sources:

• Chen, X., Wu, Z., Li, Q., Karimi, H. R., & Li, Z. (2025). A Unified Deep Reinforcement Learning Energy Management Strategy for Multi-powertrain Vehicles Based On Meta Learning and Hard Sample Mining. Control Engineering Practice, 163.
• NewsRx. (2025, October 13). Recent Findings from Beijing Institute of Technology Has Provided New Information about Technology (A Unified Deep Reinforcement Learning Energy Management Strategy for Multi-powertrain Vehicles Based On Meta Learning and Hard Sample Mining). Journal of Engineering, 2550.