Neural Network and Reinforcement Learning based Energy Management Strategy for Battery/Supercapacitor HEV

A novel energy management strategy based on neural network (NN) and reinforcement learning is proposed to split the power of lithium battery and supercapacitor for hybrid energy vehicle (HEV) driving. First, the NN state-action model is trained by using traditional energy management controller. Then, new data is generated and combined with the original data to optimize the NN model in further. The action sequences are generated randomly and evaluated by NN model. Only the first action is adopted and the next action is re-predicted to reduce the cumulative error caused by the inaccurate model. Finally, four different driving patterns are selected and the comparison with deep Q-Network method is carried out to verify the effectiveness of the proposed energy management strategy.