Energy management of hybrid electric vehicles based on model predictive control and deep reinforcement learning

In this paper, a model predictive control-deep reinforcement learning (MPC-DRL) based energy management strategy (EMS) is proposed for hybrid electric vehicles (HEVs) to realize optimal energy distribution. Firstly, the Bi-directional Long Short-Term Memory (Bi-LSTM) network is used to predict the vehicle velocity sequence over the prediction horizon. Then, a battery state-of-charge (SOC) reference trajectory planning model is constructed based on vehicle velocity changes and driving mileage. Finally, according to the predicted vehicle velocity and the SOC reference trajectory, the deep Q network (DQN) algorithm searches for optimal control solutions under the MPC framework. The simulation results show that the Bi-LSTM network can accurately predict the vehicle velocity and the MPC-DRL strategy can better follow the downward trend of the SOC reference trajectory. The proposed strategy can achieve similar fuel economy performance to the MPC-dynamic programming (DP) strategy and approximately 9.58% promotion than the ECMS strategy.