信号交叉口网联电动汽车自适应学习生态驾驶策略

Learning based eco-driving strategy of connected electric vehicle at signalized intersection

提出了一种面向信号交叉口的自适应学习生态驾驶策略。首先,搭建了电动汽车纵向动力学模型,建立了信号灯交叉路口的虚拟交通仿真环境;其次,以车辆能耗最小化与通行效率最大化为目标,耦合设计强化学习奖励函数,基于深度确定性策略梯度算法(DDPG)对车辆加速度进行实时控制与训练;最后,通过蒙特卡洛试验法,验证本文提出的强化学习生态驾驶策略在不同初始交通场景下的有效性与鲁棒性。仿真结果表明,相较于常规“加速-匀速-制动(ACB)”策略,本文提出的强化学习生态驾驶策略在单路口和多路口场景下均可有效提升通行效率和能量效率。同时,智能网联汽车数字孪生试验平台的多次实车试验表明,本文的强化学习算法控制效果良好,可以有效减少车辆路口等待时长,降低能耗同时提高通行效率。

A deep reinforcement learning based eco-driving strategy for connected electric vehicle(EV)was proposed to improve its energy efficiency at signalized intersection.Firstly,the dynamics of the EV is modelled,and the simulation environment of signalized intersection crossing scenario is established.Secondly,the reward function including multiple objectives is designed considering energy consumption reduction and travel efficiency improvement.The Deep Determinate Policy Gradient(DDPG)is developed to control the vehicle acceleration in continuous action space.Finally,a Monte Carlo simulation is conducted to verify the effectiveness and robustness of proposed method in different driving conditions.The simulation results show that the proposed strategy can improve the vehicle energy efficiency while ensuring travel efficiency in both single and multiple intersection scenarios,compared to a conventional accelerateconstant-brake strategy.In addition,a field test is conducted based on a developed connected automated vehicle digital twin platform.The experiment results show that the proposed reinforcement learning based eco-driving strategy has the potential to improve the vehicle energy efficiency and travel efficiency,simultaneously.