Reinforcement Learning-Based Energy Management for Plug-in Hybrid Electric Vehicles

In this work, two reinforcement-learning-based energy management strategies are investigated for plug-in hybrid electric vehicles (PHEVs). The strategies are developed based on the Toyota Prius model, for which the dynamics are firstly established. Then, both the deep Q-learning (DQN) and deep deterministic policy gradient (DDPG) algorithms are formulated based on the vehicle model. Furthermore, to verify the effectiveness, the reinforcement-learning-based strategies are also benchmarked against the dynamic programming under the New European Driving Cycle (NEDC). The simulation results indicate that the DQN and DDPG algorithms lag the global optimizationbased DP by around 9% and 4% in fuel efficiency, respectively.