Scholarlink

<p>In high-speed aggressive steering and low-adhesion limit conditions, tire nonlinear saturation and model parameter mismatch are critical factors compromising vehicle stability and safety. This paper focuses on distributed drive electric vehicles (DDEVs) and proposes a parameter-adaptive stability control method that integrates Multi-Head Attention-based Soft Actor-Critic (MHA-SAC) reinforcement learning with nonlinear model predictive control (NMPC). The MHA-SAC framework is designed to online adaptively adjust NMPC controller parameters in response to instantaneous variations in vehicle states under extreme maneuvers. Specifically, the MHA mechanism is employed to perform feature weighting on historical vehicle states, effectively capturing temporal dependencies. Based on these extracted features, the SAC algorithm online optimizes the NMPC prediction horizon, the cost function weighting factors, and the lower-layer torque distribution objective coefficients. Systematic validation is conducted via a Simulink/Carsim co-simulation platform under diverse scenarios, including high/low-speed slalom, fishhook, and icy double lane change (DLC) maneuvers. Simulation results demonstrate that, compared with the conventional fixed-parameter NMPC, the proposed MHA-SAC-NMPC algorithm achieves superior performance in terms of yaw rate tracking, lateral velocity suppression, and wheel slip ratio smoothness, thereby significantly extending the vehicle’s stable operating envelope under extreme conditions.</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

A parameter-adaptive stability control method for distributed electric drive vehicles under extreme operating conditions

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Yufang Li,
  • Siyu Xu,
  • Dexing Gao,
  • Tianci Zhang,
  • Xuhao Zhang,
  • Jihang Li,
  • Yuhang Wang

摘要

In high-speed aggressive steering and low-adhesion limit conditions, tire nonlinear saturation and model parameter mismatch are critical factors compromising vehicle stability and safety. This paper focuses on distributed drive electric vehicles (DDEVs) and proposes a parameter-adaptive stability control method that integrates Multi-Head Attention-based Soft Actor-Critic (MHA-SAC) reinforcement learning with nonlinear model predictive control (NMPC). The MHA-SAC framework is designed to online adaptively adjust NMPC controller parameters in response to instantaneous variations in vehicle states under extreme maneuvers. Specifically, the MHA mechanism is employed to perform feature weighting on historical vehicle states, effectively capturing temporal dependencies. Based on these extracted features, the SAC algorithm online optimizes the NMPC prediction horizon, the cost function weighting factors, and the lower-layer torque distribution objective coefficients. Systematic validation is conducted via a Simulink/Carsim co-simulation platform under diverse scenarios, including high/low-speed slalom, fishhook, and icy double lane change (DLC) maneuvers. Simulation results demonstrate that, compared with the conventional fixed-parameter NMPC, the proposed MHA-SAC-NMPC algorithm achieves superior performance in terms of yaw rate tracking, lateral velocity suppression, and wheel slip ratio smoothness, thereby significantly extending the vehicle’s stable operating envelope under extreme conditions.