Fast Learning-based Control for Energy Management of Hybrid Electric Vehicles

Abstract

In this paper, a fast learning-based control method is proposed for energy management of hy-brid electric vehicles. First, the modeling of a parallel hybrid electric vehicle (HEV) is introduced. Energy management of the parallel HEV is constructed as an optimal control problem. Then, the reinforcement learning (RL) framework is depicted and a learning-based approach named Dyna-H algorithm is illustrated via incorporating a heuristic planning strategy into a Dyna agent. Finally, the proposed energy management strategy is compared with the benchmark methods to show its merits. Results indicate that the learning-based controls have better performance in fuel economy and calculation speed.