Reinforcement Learning Energy Management for Hybrid Electric Tracked Vehicle with Deep Deterministic Policy Gradient

Reinforcement learning (RL) has been applied to energy management of hybrid electric vehicles to synthesize the system efficiency and adaptability. However, the existing RL-based energy management strategies still suffer the “curse of dimensionality” due to the discretization of the state and control action variables. To cure this disadvantage, a continuous RL-based energy management adopting deep deterministic policy gradient (DDPG) is proposed and applied to a series hybrid electric tracked vehicle. First, DDPG-based energy management strategy is put forward, where two sets of neural networks are adopted to parameterize strategy and approximate the action-value function respectively to eliminate the discretization. In addition, an online updating framework of energy management is carried out to increase the adaptability of the energy management strategy. The simulation results show that the fuel consumption of the online updating strategy is 5.9% lower than that of the stationary strategy, and is close to that of dynamic programming benchmark strategy. Besides, the computational burden is significantly reduced and can be implemented in real-time.