The distributed electric drive powertrain is one of the most promising ways to electrify and decarbonize heavy-duty vehicles. To improve the energy efficiency and driving comfort of distributed drive electric buses, this study proposes an Energy Management Strategy based on the Hybrid-Action Multi-Agent Deep Deterministic Policy Gradient algorithm (HA-MADDPG). The MADDPG algorithm enhances the EMS's adaptability and rapid response in varying environments through centralized training and decentralized execution strategy. Considering that single-motor and dual-motor driving modes are available, a rule-learning combined strategy is applied to select discrete driving modes, which adopts a look-up table to determine the single-motor mode, while continuous actions simulate the selection of the dual-motor driving mode. Expert experience is introduced in the HA-MADDPG to guide exploration and correct the unreasonable exploration by Agents, which accelerates the training speed and improves training quality. Results show that the proposed HA-MADDPG EMS improves energy efficiency by up to 1.59 % compared with the Real-time Rule-based EMS, which is widely adopted in real vehicles. Compared with theoretically Optimal Rule-based EMS, HA-MADDPG-based EMS achieves similar energy performance with fewer mode switching actions.
扫码关注我们
求助内容:
应助结果提醒方式:
