A Joint Eco-driving Optimization Research for Connected Fuel Cell Hybrid Vehicle via Deep Reinforcement Learning
With the rapid development of the new technologies about Internet of Things(IoT)and automatic driving,an advanced research target has been injected into the optimization of eco-driving and energy management of hybrid vehicles based on the connected driving environment.Aiming at the fuel cell hybrid vehicles driving on multi-signalized urban roads,this paper proposes a hierarchical multi-objective optimization method combined deep deterministic policy gradient and dynamic planning(DDPG-DP)for speed planning and energy management.The DDPG algorithm is used in the upper layer of energy-saving speed planning,while the multi-objective reward value function and the priority experience replay mechanism are designed to carry out the multi-objective speed planning for energy saving,driving comfort,and passage efficiency on the basis of improving the algorithm's speed and stability,and the dynamic planning algorithm is used in the lower layer of energy management to achieve the optimal energy-saving of the hybrid system with the goal of minimizing the hydrogen consumption.In scenarios 1 and 2,the results show that the DDPG-DP algorithm improves the traveling efficiency by 15.25％and 20.18％than the IDM-DP algorithm,and reduces the hydrogen fuel consumption by 25.66％and 17.86％,respectively.Meanwhile,there is a gap of only about 5 s in the passing time of the DDPG-DP algorithm compared with the global optimal algorithm(DP-DP)in Scenarios 1 and 2,and the hydrogen fuel consumption is lower than the optimal algorithm.Meanwhile,there is only a difference of about 5 s between the DDPG-DP algorithm and the global optimal algorithm(DP-DP)in traveling time,and there is only a difference of 2.84％and 4.7％in the hydrogen fuel consumption compared with the DP-DP algorithm.In field of driving smoothness,the DDPG-DP algorithm has less speed fluctuation than the other two algorithms(IDM-DP and DP-DP)and doesn't have large acceleration/deceleration.It will provide greater energy-saving potential for daily driving of hybrid vehicles and support the further research for multi-objective eco-driving optimization of connected fuel cell hybrid vehicles.
energy managementfuel cellhybrid vehicledeep reinforcement learningco-optimizationconnected and autonomous vehicles
万方数据
维普
