摘要
针对基于传统Q-learning的汽车协同式自适应巡航算法加减速的控制不够线性的问题,且为了进一步提高动作选择的准确性,提出一种基于LSTM改进的Q-learning控制算法。算法在Q-learning的动作选择阶段引入LSTM模型,通过输入汽车当前状态对加减速动作进行预测,能够提高Q-learning算法的控制精度,对汽车的运动控制更加稳定。通过仿真实验模拟了汽车在匀速和加减速工况下的运动控制效果,实验结果表明,提出的算法与传统Q-learning和Deep Q-learning算法相比,跟车速度和跟车距离的平均误差和平均标准差明显更小,速度控制曲线更加平滑。
Aiming at the problem that the acceleration and deceleration control of the traditional Q-learning based on the automotive cooperative adaptive cruise algorithm is not linear enough,and to further improve the accuracy of action selection,so an improved Q-learning control algorithm based on LSTM is proposed in this paper.This algorithm introduces the LSTM model in the action selection stage of Q-learning,and predicts the acceleration and deceleration actions by inputting the current state of the vehicle,which can improve the control accuracy of Q-learning and make the motion control of the vehicle more stable.This paper simulates the motion control effect of the car under constant speed and acceleration/deceleration conditions through simulation experiments.The experiment results show that the proposed algorithm has a smoother speed control curve compared with the traditional Q-learning and Deep Q-learning algorithm,and the mean error and mean standard deviation of following speed and following distance are smaller obviously.
作者
吴金
WU Jin(School of Information and Intelligent Transportation,Fujian Chuanzheng Communications College,Fuzhou 350007,China)
出处
《长春工程学院学报（自然科学版）》
2020年第4期65-70,共6页
Journal of Changchun Institute of Technology：Natural Sciences Edition
基金
福建省教育厅中青年教师教育科研项目(JAT191204)。
作者简介
吴金(1983-),男(汉),福州人,硕士,讲师主要研究计算机科学与应用。
