Dual-layer adaptive power management strategy for E-tractor incorporating operating information and deep reinforcement learning

Electric tractors are crucial for sustainable agriculture, but optimizing their power management strategies (PMS) to achieve energy savings remains challenging. This paper proposes an adaptive power distribution method of hybrid energy storage system (HESS) for electric tractors to further reduce battery capacity degradation and power loss costs. Firstly, a dataset based on the tractor's actual plowing operating conditions was created. Then, an operation condition recognition (OCR) method was designed for electric tractors by capturing the characteristic parameters of plowing operation information. Subsequently, a dual-layer adaptive PMS, termed OCR-SAC PMS, is established by integrating OCR and deep reinforcement learning (DRL) within a comprehensive optimization framework focused on minimizing operating costs. The Soft Actor-Critic (SAC) algorithm is utilized to continuously optimize power allocation. Simulation and experimental results demonstrate that the OCR achieves a recognition accuracy of 98 %. Furthermore, the proposed OCR-SAC PMS reduces operating costs by over 13 %, effectively suppresses battery peak power transients by more than 11.61 %, and reduces peak current transients by over 17.14 % compared to conventional PMSs. Additionally, optimizing the PMS through OCR results in a 70 % reduction in SAC training time.