An efficient intelligent energy management strategy based on deep reinforcement learning for hybrid electric flying car

Hybrid electric flying cars hold clear potential to support high mobility and environmentally friendly transportation. For hybrid electric flying cars, overall performance and efficiency highly depend on the coordination of the electrical and fuel systems under ground and air dual-mode. However, the huge differences in the scale and fluctuation characteristics of energy demand between ground driving and air flight modes make the efficient control of energy flow more complex. Thus, designing a power coordinated control strategy for hybrid electric flying cars is a challenging technical problem. This paper proposed a deep reinforcement learning-based energy management strategy (EMS) for a series hybrid electric flying car. A mathematical model of the series hybrid electric flying car driven by the distributed hybrid electric propulsion system (HEPS) which mainly consists of battery packs, twin turboshaft engine and generator sets (TGSs), 16 rotor-motors, and 4 wheel-motors is established. Subsequently, a Double Deep Q Network (DDQN)-based EMS considering ground and air dual driving mode is proposed. A simplified method for the number of control variables is designed to improve exploration efficiency and accelerate the convergence speed. In addition, the frequent engine on/off problem is also taken into account. Finally, DDQN-based and dynamic programming (DP)-based EMSs are applied to investigate the power flow distribution for two completely different hypothetical driving scenarios, namely search and rescue (SAR) scenarios and urban air mobility (UAM) scenarios. The results demonstrate the effectiveness of the DDQN-based EMS and its capacity of reducing the computation time.