Parameterized deep Q-network based energy management with balanced energy economy and battery life for hybrid electric vehicles

Energy management strategy (EMS) is an essential technique to ensure the long- term driving economy of hybrid electric vehicles (HEVs). The complicated discrete–continuous hybrid action space lying in HEV's driving system presents a challenge to achieve high-performance EMSs. Thus, this paper proposes a novel improved deep Q-network (DQN)-based EMS to reduce the HEV's driving costs, with lithium-ion battery (LIB) life and energy economy considered. Firstly, a data-driven battery life map reflecting the non-linear decaying trajectory of battery state of health (SOH) is proposed to quantify the real-time battery aging. Secondly, in the proposed EMS incorporating the battery aging model, an enhanced parameterized DQN (PDQN) algorithm is applied to particularly provide a hybrid solution discriminating between discrete and continuous actions. Finally, with the dynamic programming (DP) method employed as the benchmark, the effectiveness and optimality of the proposed EMS are validated. Without the prior knowledge of testing driving conditions, the proposed EMS effectively achieves 99.5% performance of the DP method, reducing the vehicle's driving costs by 3.1% and extending battery life effectively. The EMS converges quickly during training and a hardware-in-loop test validates its real application potential.