Online Updating Energy Management Strategy Based on Deep Reinforcement Learning with Accelerated Training for Hybrid Electric Tracked Vehicles

An online updating energy management strategy (EMS) based on deep reinforcement learning (DRL) with accelerated training is proposed to further reduce fuel consumption and improve the adaptability of the algorithm. The online frame continuously updates neural network parameters every predetermined time. First, the mathematical model of a series hybrid electric tracked vehicle (SHETV) is established, and the accuracy of the model is verified by collecting data from the real vehicle. Second, an intelligent EMS is developed by combining deep deterministic policy gradient (DDPG) and prioritized experience replay (PER). DDPG can improve the control effect and accelerate the training process by eliminating the discretization of variables. The addition of PER can further improve fuel economy and SOC performance and shorten the training time. Third, an online updating framework based on DDPG-PER is proposed to make EMS better adapt to complex driving conditions. Finally, a software-in-the-loop simulation is built, and the effectiveness of the proposed online updating EMS is verified by comparison with other state-of-the-art algorithms. Simulation results show that the training process of the off-line EMS is greatly accelerated, which provides a potential for online application. Meanwhile, the fuel economy of the online updating EMS reached 93.9% of the benchmark DP.