Energy Management Strategy Based on an Improved TD3 Reinforcement Algorithm with Novel Experience Replay

The energy management strategy (EMS) plays an important part in the systematic control of hybrid electric vehicles (HEVs). In recent years, the EMS based on deep reinforcement learning (DRL) receives more attention. This paper proposes an EMS based on TD3 deep reinforcement learning algorithm with novel experience replay. The experience replay is introduced to select samples via an evaluation network aiming to improve the learning ability and the convergence speed. The results show that compared with the traditional TD3-based EMS, the proposed EMS reduces the training time by 8.73% and improves the fuel economy by 2.14%.

I. Introduction

In the face of the energy crisis and climate change, there is a huge transformation in the automobile industry, and hybrid electric vehicle (HEV) is regarded as a feasible technical solution [1]. HEVs allocated the power properly between the engine and motor to combine the merits of fuel economy and system reliability [2].