Improved deep reinforcement learning optimized energy efficiency management strategy of hybrid electric commercial vehicles based on operational & maintenance costs

As profit-driven production assets, commercial vehicles are more reasonable to adapt decision objectives based on full-life-cycle operational and maintenance (O&M) costs for energy efficiency management strategies (EEMS). This study proposes an improved deep reinforcement learning (DRL) optimized EEMS for hybrid electric commercial vehicles (HECVs). First, the O&M cost framework for the fuel system comprises engines and fuel system components, whereas the electrical system encompasses motors, power batteries, and electronic control units, based on actual repair and replacement data. Second, the soft actor–critic (SAC) DRL is employed to explore the EMMS for HECVs, enhancing optimization via an unguided reward function architecture. Finally, the control performances under different EEMS are analyzed using real driving data, incorporating real-time road gradients for training and testing. The results indicated that compared to strategies based on fuel economy, the strategies targeting O&M costs increase fuel costs by 15.51% and emission treatment (ET) costs by 21.15%. However, engine maintenance and electrical system costs decrease by 39.59%, resulting in an overall O&M cost reduction of 22.24%. This demonstrates that decision objectives centered on O&M costs better align with the operational characteristics of HECVs. Additionally, the DRL strategies employing an unguided reward function consistently achieve lower total O&M costs than traditional reward function approaches, demonstrating the superiority of the unguided reward function architecture. Furthermore, the proposed strategy achieved a dynamic programming (DP) performance of 88.16% while maintaining real-time control capability and delaying electronic system degradation.