Scholarlink

<p>As a data-driven modeling approach, the Koopman operator theory captures nonlinear characteristics of electromechanical systems (e.g., vehicles) more accurately than physics-based methods. In this paper, we employ the Extended Dynamic Mode Decomposition (EDMD) algorithm to extract basis functions for the Koopman operator, mapping the nonlinear vehicle system dynamics into a linear representation in a lifted space. The selection of basis functions for state lifting is crucial in Koopman-based model identification. Therefore, we conduct a comparative study of several typical basis functions and investigate the impact of basis function dimensionality on model accuracy. Additionally, a Moving Horizon Estimation (MHE) model is introduced to accurately predict lateral vehicle velocity when direct measurement is challenging. The proposed model is then applied to predict vehicle states over a prediction horizon, forming a constrained finite-time optimal control problem under the Model Predictive Control (MPC) framework, termed MHE-KMPC. Simulation results under three lane-change scenarios using a CarSim/Simulink co-simulation platform demonstrate that the proposed MHE-KMPC achieves higher tracking accuracy and enhanced real-time performance compared to both linear and nonlinear MPC benchmarks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Incorporating Moving Horizon Estimation into Koopman Model Predictive Control for Autonomous Vehicle Path Tracking

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Gan Shen,
  • Bingli Zhang,
  • Yixin Wang,
  • Chengbiao Zhang,
  • Xiang Luo,
  • Xinyu Wang,
  • Zhongzheng Liu,
  • Da Zheng,
  • Yunsheng Ye

摘要

As a data-driven modeling approach, the Koopman operator theory captures nonlinear characteristics of electromechanical systems (e.g., vehicles) more accurately than physics-based methods. In this paper, we employ the Extended Dynamic Mode Decomposition (EDMD) algorithm to extract basis functions for the Koopman operator, mapping the nonlinear vehicle system dynamics into a linear representation in a lifted space. The selection of basis functions for state lifting is crucial in Koopman-based model identification. Therefore, we conduct a comparative study of several typical basis functions and investigate the impact of basis function dimensionality on model accuracy. Additionally, a Moving Horizon Estimation (MHE) model is introduced to accurately predict lateral vehicle velocity when direct measurement is challenging. The proposed model is then applied to predict vehicle states over a prediction horizon, forming a constrained finite-time optimal control problem under the Model Predictive Control (MPC) framework, termed MHE-KMPC. Simulation results under three lane-change scenarios using a CarSim/Simulink co-simulation platform demonstrate that the proposed MHE-KMPC achieves higher tracking accuracy and enhanced real-time performance compared to both linear and nonlinear MPC benchmarks.