Energy Optimization of Hybrid electric Vehicles Using Deep Q-Network

Hybrid electric vehicles are positioned as an intermediate form between gasoline and electric vehicles, contributing to lower fuel consumption and emissions. Map control is used for conventional engine control. This method maps optimal values from experimental data, and it has been pointed out that the capacity of the map is increasing and that it is difficult to respond to increasingly sophisticated control objectives. In this paper, we first present a model of a series-parallel hybrid electric vehicle and propose a method using Deep Q-Network, a typical reinforcement learning technique. Through numerical simulation, we verify that the SOC is within an acceptable range throughout the entire run and that energy efficiency can be improved compared to existing map control.