Deep Reinforcement Learning-based Control Strategy of Connected Hybrid Electric Vehicles Platooning
In view of the characteristics of strong nonlinearity and hybrid driving with multi-power sources,a deep learning based hierarchical control strategy of connected hybrid electric vehicles platooning is proposed.Firstly,a model predictive controller for platoon is designed to solve the expected optimal and multi-objective acceleration of vehicles based on the vehicular state acquiring by vehicle to vehicle communication.Secondly,the optimal operating curve of the engine and the battery characteristic curve are embedded into the deep learning algorithm as the expert knowledge.Then,the influence of battery power,engine power,vehicle speeds and vehicle accelerations on the action value of the agent is discussed to illustrate how the DQN deep learning-based energy management control algorithm realizes the coordinated control of the multi-system power output of vehicles in the platoon according to the action value.Finally,a reward function with battery state of charge and instantaneous fuel consumption as independent variables is designed.By using the minimization loss function,the parameters of the DQN network are updated by the gradient descending method,and the energy management control is realized by the deep reinforcement learning method.The test results show that the proposed control strategy can dynamically plan the expected acceleration of vehicles in the platoon,and reasonably allocate between the engine power and motor power in real time,and finally achieve energy-saving driving of vehicles in the platoon.
connected hybrid electric vehiclesplatoondeep reinforcement learningmodel predictive controlenergy-saving driving
NETL
NSTL
万方数据
维普
