Research On Eco-driving Strategy Of Hybrid Electric Truck Based On Reinforcement Learning

In the past thirty years,people’s emphasis on fossil fuel consumption and environmental pollution has completely changed the current automotive industry.New energy technologies such as Hybrid Electric Vehicle(HEV)are receiving attention.The HEV technology route not only solves the range anxiety of pure electric vehicles,but also reduces energy consumption and pollutant emissions to some extent,and has many advantages in fuel economy and other aspects.Eco-driving strategy is one of the key technologies for energy conservation and emission reduction in new energy vehicles.It can optimize and control vehicle energy conservation from multiple dimensions such as route layer,speed layer,and power layer,enabling vehicles to drive in a safe,efficient,and energy-saving manner.This paper is dedicated to developing an eco-driving strategy based on reinforcement learning for a series-parallel hybrid electric truck.The strategy is divided into two layers,which are the upper layer for adaptive cruise control(ACC)strategy and the lower layer for energy management strategy(EMS).For the cruise control strategy of the upper layer,this study uses the deep deterministic policy gradient(DDPG)algorithm to develop the eco-driving cruise control strategy for the hybrid light truck driving on the fixed line.Considering the development status of the Internet of Vehicles technology,this strategy is applicable to the situation where the traffic status data is relatively small,and only provides the traffic light phase information and limited sensor data to the control algorithm.In order to verify the efficiency and energy saving advantages of the strategy proposed in this study,the GLOSA cruise strategy included in the traffic simulation software is used as a reference.The simulation results show that compared with the suboptimal GLOSA strategy with similar travel time,the energy consumption of the DDPG-based cruise strategy proposed in this study can be reduced by about 10%,and the cruise safety can be guaranteed.For the energy management strategy at the lower layer,series-parallel P2-P3 HEV have complex topologies with multiple power sources and multiple working modes,which make the development of effective EMS challenging.In this paper,a deep reinforcement learning(DRL)combined with pre-optimization energy management framework is proposed.Incorporating the characteristics of this HEV,this study embedded the pre-optimized equivalent motor and pre-optimized mode selection rule into a DDPG loop.The pre-optimization of the equivalent motor aims to equate two motors as the equivalent motor,thereby reducing the dimensionality of the energy management task,and the pre-optimized mode selection rule aims to systematically integrate mode selection and energy allocation into the EMS.In this study,it is assumed that HEV operate in connected urban environments.This study introduce middle-horizon traffic information to improve the performance of EMS.Furthermore,this study also consider the frequent engine start-stop characteristics of HEV in EMS design.The simulation results show that the fuel economy of several EMS proposed in this study can reach 87.2% to 90.7% of Dynamic Programming(DP)based EMS.The introduction of an engine start-stop penalty can significantly reduce the number of engine start-stops without sacrificing fuel economy.