汽车工程 ›› 2023, Vol. 45 ›› Issue (12): 2260-2271.doi: 10.19562/j.chinasae.qcgc.2023.12.008

所属专题: 底盘&动力学&整车性能专题2023年

• • 上一篇    下一篇

基于Q学习的整车主动悬架免参数H控制

王刚,李昆鹏,景晖(),刘溯奇   

  1. 桂林电子科技大学,广西制造系统与先进制造技术重点实验室,桂林 541004
  • 收稿日期:2023-04-11 修回日期:2023-06-02 出版日期:2023-12-25 发布日期:2023-12-21
  • 通讯作者: 景晖 E-mail:jinghuiedu@163.com
  • 基金资助:
    国家自然科学基金(12202112);广西自然科学基金(2021JJB160015);广西制造系统与先进制造技术重点实验室主任项目(22-35-4-S006)

Parameter-Free H Control of Vehicle Active Suspension Based on Q-learning

Gang Wang,Kunpeng Li,Hui Jing(),Suqi Liu   

  1. Guilin University of Electronic Technology,Guangxi Key Laboratory of Manufacturing System andAdvanced Manufacturing Technology,Guilin  541004
  • Received:2023-04-11 Revised:2023-06-02 Online:2023-12-25 Published:2023-12-21
  • Contact: Hui Jing E-mail:jinghuiedu@163.com

摘要:

主动悬架是智能汽车全线控底盘的重要部件,配合各线控执行系统可实现整车底盘的全矢量控制,极大增强行驶的安全性,而传统控制方法需要标定的整车模型参数过多,降低了控制开发的效率。基于此,本文研究了整车主动悬架免参数H控制方法。首先,建立主动悬架的行为依赖近似动态规划模型,将H控制问题转化为路面干扰和控制行为的零和博弈过程;其次,使用自适应评判方法整定动作网络与批判网络,通过在线Q学习求解系统的博弈黎卡提方程,给出无须模型参数的控制最优解,稳定性分析表明该方法可收敛到系统的纳什平衡点;最后,搭建硬件在环系统验证该方法的有效性,对包块路面以及不同路面等级下的随机路面进行主动控制仿真。结果表明,基于Q学习的控制方法具有最优控制效果,能够改善低频范围内的整车平顺性和操纵稳定性。

关键词: 整车主动悬架, 免参数, H控制, Q学习

Abstract:

Active suspension plays a crucial role in the all-by-wire chassis of intelligent vehicles, enabling full-vector control of the vehicle chassis in conjunction with various driven-by-wire execution systems, significantly improving driving safety. However, the traditional control method involves calibrating numerous vehicle model parameters, which reduces the efficiency of control development. On this basis, a parameter-free H control method for the vehicle's active suspension is studied in this paper. Firstly, an approximate dynamic programming model based on the behavior dependence of the active suspension is established, transforming the H control problem into a zero-sum game process involving pavement disturbance and control behavior. Secondly, an adaptive evaluation method is used to set the action network and the critical network and online Q-learning is used to solve the system's Game algebraic Riccati equation, which provides an optimal control solution without requiring model parameters. The stability analysis demonstrates that the proposed method can converge to the Nash equilibrium point of the system. Finally, a hardware-in-the-loop system is built to verify the effectiveness of the proposed method, and the active control simulation is carried out for the bump pavement and random pavement with different road grades. The results show that the control method based on Q-learning has the optimal control effect, which can improve vehicle ride comfort and handling stability in the low frequency range.

Key words: full-car active suspension, parameter-free, H control, Q-learning