Terrain information-involved power allocation optimization for fuel cell/battery/ultracapacitor hybrid electric vehicles via an improved deep reinforcement learning

Abstract

As for fuel cell hybrid electric vehicles (FCHEV) powered by fuel cell, battery, and ultracapacitor, its complex topology structure and the variable terrain environment bring challenges to the energy management for power sources lifespan and fuel economy. In this paper, a terrain information-involved energy management strategy (EMS) obtained by an improved deep reinforcement learning (DRL) algorithm is proposed to distribute reasonably the power demand for FCHEV. Specially, an action-value-based adaptive noise and an action-screening-mechanism-based priority experience replay are designed for efficient strategy learning with the adopted algorithm. To protect fuel cell and battery from peak power, a hierarchical energy management framework is established by introducing an adaptive fuzzy filter. Then, an equivalent consumption minimization strategy based on multi-objective optimization is constructed and used as the reward of the improved algorithm to achieve the minimum fuel consumption. Meanwhile, based on historical experimental data, the terrain-information-considered transition probability matrix is obtained and applied to the algorithm to reduce computational load and get the optimal EMS. Finally, the effectiveness of the proposed EMS is verified by a series of comprehensive simulations under different driving cycles. The simulation results show that the hierarchical EMS framework can better protect the service life of fuel cell and battery. In addition, compared with the EMS without considering terrain information, the incorporation of terrain information in EMS can enhance the operational efficiency of fuel cell, improve fuel economy by about 8%, and can reach the dynamic programming benchmark level of 89.5%.