Energy Management Strategy Of Plug-in Hybrid Electric Bus Based On Driving Style And Deep Reinforcement Learning

Environmental pollution and resource shortage are becoming more and more serious.New energy vehicles have become an effective way to solve these problems.Plug-in hybrid electric buses（PHEBs）combine the advantages of conventional fuel vehicles and pure electric vehicles which can achieve a longer mileage and less emissions and are widely used in urban passenger transport.One of the key technologies in PHEV is the energy management strategy（EMS）,which directly affects the performance of PHEV by controlling the energy flow between the engine and the motor to achieve optimal torque/power distribution.The energy management of PHEV is synthetically affected by human-vehicle-road.This paper focuses on the research of PHEV energy management considering the influence of driving cycles and driver’s driving style.In this paper,the bus prototype studied is a PHEB of Zhongtong bus company,and its hybrid power system adopts a single shaft parallel structure.Firstly,the mathematical models of the engine,motor,power battery and other key components as well as the vehicle dynamics system were established.The vehicle simulation model was built based on AVL CRUISE simulation platform,and the vehicle dynamic performance was verified.Based on the established vehicle model,two optimization-based EMSs were designed,including real-time optimization EMS based on equivalent fuel consumption（ECMS）and global optimization EMS based on dynamic programming（DP）.They are compared with charge decoupling-charge sustaining（CD-CS）strategy to verify its effectiveness and provide evaluation criteria for the following EMSs.Three advanced deep reinforcement learning（DRL）algorithms was introduced in detail,including double deep Q network（DDQN）,deep deterministic strategy gradient（DDPG）and twin delayed deep deterministic policy gradient（TD3）.Then,the EMS based on DRL was designed by taking the demand torque,vehicle speed,acceleration and battery SOC as the system state,the engine output torque as the control action,and the fuel consumption and battery SOC stability as the reward.The simulation results show that TD3 based energy management has faster convergence speed,stronger selflearning ability and better fuel economy.On the basis of the above research,considering the impact of driving style on vehicle fuel economy,the DRL-based EMS considering driving style is proposed.The influence of driving cycles on driving style was analyzed,and two typical driving cycles were selected as standard driving cycles.And the driving cycles were recognized by principal component analysis（PCA）and cluster analysis.On this basis,drivers’ driving styles was divided into calm,moderate and aggressive.Then driving style recognition was combined with CD-CS strategy,DP strategy and DRL-based strategy to complete the design of energy management strategy considering driving style,and the simulation was verified under different driving cycles.The results show that compared with the TD3-based strategy without considering driving style,the fuel economy of the TD3based strategy considering driving style recognition under real driving cycle and typical combined driving cycle is improved by 7.84%and 14.71%respectively.This shows that combining driving style recognition with DRL-based EMS can effectively improve the fuel economy of PHEB.