An Improved Deep Reinforcement Learning Based Energy Management Strategy for Hybrid Electric Agricultural Tractor

Current hybrid tractors predominantly employ traditional rules-based energy management strategy, which can be inefficient and heavily reliant on professional expertise. Additionally, the working cycle for these tractors mainly utilizes the existing vehicle road cycle, which is inadequate in meeting the working conditions of tractors, thereby resulting in the underutilization of hybrid tractors for their energy-saving potential. In this paper, a novel tractor working cycle is constructed, which takes into consideration the power take-off mechanism of the hybrid tractor and is, therefore, more compatible with tractor working conditions. Moreover, an improved deep reinforcement learning-based energy management strategy is proposed, named softmax deep deterministic policy gradient, based on the newly constructed working cycle to avoid inefficient searching, thereby enhancing the algorithm’s efficiency.