Control of Hybrid Electric Vehicle Powertrain Using Offline-Online Hybrid Reinforcement Learning

Hybrid electric vehicles can achieve better fuel economy than conventional by utilizing multiple power sources. While these sources have been controlled rule-based or optimization-based control algorithms, recent studies shown that machine learning-based algorithms such as online Deep Reinforcement Learning (DRL) effectively the well. However, optimization and training processes for DRL-based powertrain strategy be very time resource intensive. In this paper, a new offline–online hybrid DRL is presented where offline vehicle data are exploited to build an initial model learning algorithm explores policy further improve economy. manner, it expected agent learn environment consisting of dynamics in given driving condition more quickly compared which optimal interacting with from zero knowledge. By incorporating priori knowledge, simulation results show proposed approach not only accelerates process makes stable, but also leads algorithms.