• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2024, Vol. 60 ›› Issue (10): 453-462.doi: 10.3901/JME.2024.10.453

• 先进控制技术 • 上一篇    下一篇

扫码分享

基于自学习型MPC的网联电动汽车生态驾驶控制策略研究

李兵兵1, 庄伟超1, 刘昊吉1, 张昊2, 殷国栋1, 张建润1   

  1. 1. 东南大学机械工程学院 南京 211189;
    2. 清华大学汽车安全与节能国家重点实验室 北京 100084
  • 收稿日期:2023-07-18 修回日期:2024-01-07 出版日期:2024-05-20 发布日期:2024-07-24
  • 作者简介:李兵兵,男,1994年出生,博士研究生。主要研究方向为智能网联汽车、经济性驾驶和车辆动力学控制。
    E-mail:bingbli@seu.edu.cn
    庄伟超(通信作者),男,1990年出生,博士,副教授,博士研究生导师。主要研究方向为智能汽车自主决策与网联协同控制、经济性驾驶及能量管理。
    E-mail:wezhuang@seu.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(52172383)。

Research on Eco-driving Control Strategy of Connected Electric Vehicle Based on Learning-MPC

LI Bingbing1, ZHUANG Weichao1, LIU Haoji1, ZHANG Hao2, YIN Guodong1, ZHANG Jianrun1   

  1. 1. School of Mechanical Engineering, Southeast University, Nanjing 211189;
    2. State Key Laboratory of Automotive Safety and Energy, Tsinghua University, Beijing 100084
  • Received:2023-07-18 Revised:2024-01-07 Online:2024-05-20 Published:2024-07-24

摘要: 生态驾驶是实现可持续出行和可持续城市交通发展的重要途径。为提升网联电动汽车的能量效率,针对复杂多变的城市信号灯路口场景,充分考虑真实交通的信号灯配时以及车辆对未来信息的有限预测能力等约束条件,提出一种结合学习型模型预测控制(Learning-based model predictive control, Learning-MPC)与快速内点法(Fast internal point method,FIPM)的两阶段非保守生态驾驶控制策略(Non-conservative ecological driving control, NCEDC)。车辆出发前,根据乘客目的地以及道路限速信息,构建能效最优控制问题,同时为提高计算效率引入带阻函数将速度约束转化为目标函数一部分,内点法粗规划车辆能效最优参考速度轨迹;出发时,车辆预测信号灯动态相位,Learning-MPC策略通过高斯过程(Gaussian process,GP)在线学习车辆状态预测模型,实现车辆能效最优参考速度轨迹的跟踪控制。通过仿真对比,相对于经典的“加速—匀速—减速”策略,所提方法可实现9.7%的能量节省,并随着预测视域的长度增加表现出更好的节能潜力。进一步验证通过机器学习解决传统MPC非柔性保守系统状态预测模型因离散化造成的误差累积问题,更高程度提升了车辆生态驾驶控制的最优效果。

关键词: 网联汽车, 生态驾驶, 模型预测控制, 高斯过程, 信号灯路口

Abstract: Eco-driving is an important way to achieve sustainable mobility and sustainable urban transport development. To improve the energy efficiency of connected electric vehicles, a two-stage non-conservative eco-driving control strategy combining learning-based model predictive control and fast interior point method is proposed for complex and variable urban signalized intersection scenarios, taking into full consideration constraints such as signal phase and timing information of real traffic and the limited predictive capability of vehicles for future information. Before vehicle departure, the energy-efficient optimal control problem is constructed based on passenger destination and road speed limit information, while the band-stop function is introduced to improve the computational efficiency to transform the speed constraint into a part of the objective function, and the interior point method coarse planning solves the vehicle energy-efficient optimal reference speed trajectory; At departure, the vehicle predicts the dynamic signal phase and timing information, and the Learning-MPC strategy learns the state prediction model of the vehicle online through Gaussian process to realize the tracking control of the vehicle energy-efficient optimal reference speed trajectory. The simulation shows that the proposed method can achieve 9.7% energy saving compared with the classical acceleration-uniformity-deceleration strategy, and indicates better energy saving potential as the length of the predicted field of view increases. It is further verified that the error accumulation problem caused by the discretization of the traditional MPC non-flexible conservative system state prediction model is solved by machine learning, and the optimal effect of vehicle eco-driving control is improved to a higher degree.

Key words: connected vehicle, eco-driving, model predictive control, Gaussian process, signalized intersections

中图分类号: