A novel transferable energy management framework for hybrid electric tractors based on deep transfer reinforcement learning

Deep reinforcement learning (DRL)-based energy management strategies (EMSs) have shown great potential in improving energy efficiency and fuel economy of hybrid electric vehicles (HEVs) under complex operating conditions. However, most DRL-based EMSs exhibit limited generalization capability and require time-consuming retraining when driving cycles change. Integrating transfer learning (TL) with DRL offers an effective way to transfer prior knowledge to the other new domain, enabling faster convergence and improved initial performance. While existing studies mainly focus on on-road HEVs, applications to hybrid electric tractors (HETs) remain limited. Compared with HEVs, HETs face additional challenges due to strong power coupling between propulsion and power take-off (PTO) systems and highly variable working scenarios. To address these issues, this paper proposes a novel transferable DRL-based EMS for HETs, which demonstrates improved generalization and training efficiency under diverse operating conditions.