Twin delayed deep deterministic policy gradient based energy management strategy for fuel cell/battery/ultracapacitor hybrid electric vehicles considering predicted terrain information

Abstract

For fuel cell/battery/ultracapacitor hybrid electric vehicles (FCHEV), the complex topology and terrain information pose challenges to fuel cell efficiency, power sources lifespan, and fuel economy. In this paper, a hierarchical energy management strategy (EMS) considering predicted terrain information is constructed to optimize the demand power allocation of FCHEV based on twin delayed deep deterministic policy gradient (TD3). Firstly, to improve the fuel economy of FCHEV, an adaptive fuzzy filter-based upper-level frequency separation is designed considering negative power. Secondly, to achieve the minimum fuel consumption, a TD3-based lower-level EMS is designed, where the terrain information is predicted through multi long short-term memory network and considered in the TD3 framework. The optimal state of charge obtained by dynamic programming (DP) based on the demand power and predicted terrain information, is considered in the reward function of TD3, as well as the optimal experience solved by DP based on historical road data is integrated into the designed hybrid experience replay. Finally, the effectiveness and optimality of the TD3-based EMS considering predicted terrain information are verified by a series of comprehensive simulations under different driving cycles. The simulation results under CYC_SC03 and test driving cycles indicate that compared to traditional TD3-based EMS, the fuel economy is increased by 33.3% and 27.5%, respectively, and can reach the benchmark level of 80% and 80.9% of DP, respectively.