欢迎访问《汽车安全与节能学报》，
汽车安全与节能学报

汽车安全与节能学报 ›› 2025, Vol. 16 ›› Issue (4): 638-647.DOI: 10.3969/j.issn.1674-8484.2025.04.014

• 智能驾驶与智慧交通 • 上一篇    下一篇

基于神经网络自适应MPC智能车辆轨迹跟踪仿真

王琳1,2(), 陈清华3, 业红玲1, 王鹏飞1, 徐驰1,2, 钱爱文1   

  1. 1 蚌埠学院 机械与车辆工程学院， 蚌埠 233030， 中国
    2 安徽省增材制造工程研究中心， 蚌埠 233030， 中国
    3 矿山智能装备与技术安徽省重点实验室， 淮南 232001， 中国
  • 收稿日期:2025-02-18 修回日期:2025-04-10 出版日期:2025-08-30 发布日期:2025-08-27
  • 作者简介:王琳（1987—），女（汉），河南，讲师。E-mail：bbxybysj@126.com。
  • 基金资助:
    安徽省高等学校科学研究重点项目(2023AH052931);高校产学研合作项目(2024340306000238)

Simulation of intelligent vehicle trajectory tracking based on neural network adaptive MPC

WANG Lin1,2(), CHEN Qinghua3, YE Hongling1, WANG Pengfei1, XU Chi1,2, QIAN Aiwen1   

  1. 1 School of Mechanical and Automotive Engineering, Bengbu University, Bengbu 233030, China
    2 Anhui Province Additive Manufacturing Engineering Research Center, Bengbu 233030, China
    3 Anhui Provincial Key Lab of Mine Intelligent Equipment and Technology, Anhui University of Science and Technology, Huainan 232001, China
  • Received:2025-02-18 Revised:2025-04-10 Online:2025-08-30 Published:2025-08-27

RichHTML

53

PDF

138

摘要：

传统模型预测控制（MPC）控制器的权重矩阵通常依赖人工经验调参，难以适应复杂动态环境，因此，提出一种基于反向传播（BP）神经网络的MPC权重矩阵自适应调整的方法。建立MPC智能车辆动力学模型分析不同权重系数对车辆轨迹跟踪性能的影响，构造数据训练BP神经网络模型，利用Matlab/Simulink搭建BP神经网络自适应MPC控制器与Carsim联合仿真，最后从不同车速和路面附着系数2个方面设计双移线仿真工况，验证控制器在不同工况下的鲁棒性。结果表明：BP神经网络自适应MPC控制器，在路面附着系数为0.85时，不同车速下的控制效果良好；而定权重MPC控制的车辆，当车速达到65 km/h时，车辆接近失稳；前者横向位移偏差和横摆角偏差的均方根分别降低44.17%和66.66%；在不同附着系数的路面上前者表现亦佳，尤其是在附着系数0.35的湿滑路面，车速30 km/h时，相对定权重MPC控制器，两项偏差均方根分别降低27.49%和49.54%。该神经网络自适应调整MPC控制器权重的方法，可为智能网联车辆中高速协同控制和特种作业车辆自主导航的轨迹跟踪性能改善提供一定参考。

关键词: 智能网联车辆, 神经网络, 自适应, 轨迹跟踪, 模型预测控制（MPC）

Abstract:

The weight matrix of traditional model predictive control (MPC) controllers usually relies on manual experience for parameter tuning, making it difficult to adapt to complex dynamic environments. Therefore, a method for adaptive adjustment of MPC weight matrices based on backpropagation (BP) neural networks was proposed. Firstly, the intelligent vehicle dynamics model with MPC control was established to analyze the influence of different weight coefficients on the vehicle trajectory tracking performance, secondly the data were constructed to train the BP neural network model, and the BP neural network adaptive MPC controller was constructed using the Matlab/Simulink module to jointly simulate with Carsim, and finally, a double-shift simulation condition was designed from different speeds and road adhesion coefficients to validate the robustness of the controller under different working conditions. The results show that the BP neural network-based adaptive MPC controller achieves favorable control performance across different speeds when the road surface adhesion coefficient is 0.85. At a speed of 65 km/h, the vehicle under the fixed-weight MPC control approaches destabilization, whereas the root-mean-squares (RMS) of the lateral displacement deviation and lateral angle deviation for the adaptive controller are reduced by 44.17% and 66.66%, respectively. The proposed controller also exhibits strong performance on road surfaces with varying adhesion coefficients—most notably on slippery roads with an adhesion coefficient of 0.35. When traveling at 30 km/h under such conditions, the RMS values of the two deviations are decreased by 27.49% and 49.54% compared to the fixed-weight MPC controller. This neural network-based approach for adaptive adjustment of MPC controller weights can provide valuable insights for enhancing trajectory tracking performance in medium-and high-speed cooperative control of intelligent connected vehicles, as well as in autonomous navigation systems for special operation vehicles.

Key words: intelligent connected vehicle, neural network, adaptive, trajectory tacking, model predictive control (MPC)

中图分类号: