【摘要】 针对分布式驱动电动汽车在转向变道过程中低附着、高速等极限工况下的失稳问题，提出一种基于K-means++算法划分车辆状态区域的分层协同控制策略。基于Carsim车辆模型构建离线训练数据集，提取横摆角速度、质心侧偏角等9维车辆稳定性特征参数，利用K-means++算法将车辆当前状态划分为稳定域、协调域与控制域，并设计动态权重协调模块。在上层控制器中，采用离散滑模控制算法结合粒子群优化趋近律系数，生成目标附加横摆力矩，以跟踪理想横摆动力学特性；同时通过对比积分滑模算法，验证离散滑模控制器在抑制峰值误差与跟踪精度上的优势。在下层控制器中，以稳定性裕度建立目标函数，构建二次规划模型，优化四轮扭矩分配，确保纵向力与侧向力矢量位于摩擦椭圆内。Carsim/Simulink联合仿真验证表明：该策略在中速、低附着(60 km/h,μ=0.3)工况下，相较于由积分滑模算法所搭建的控制策略而言，横摆角速度、质心侧偏角的峰值误差分别降低了77.2%、11.64%,而在跟踪精度方面分别优化了63.13%、15.19%;在高速、高附着(95 km/h,μ=0.85)工况下，其横摆角速度、质心侧偏角的峰值误差分别降低了27.48%、40.1%,而在跟踪精度方面分别优化了20.67%、45.94%。研究结果表明：基于K-means++算法的状态区域划分与离散滑模分层动态控制机制显著提升了车辆横向稳定性与控制鲁棒性，为分布式驱动电动汽车的极限工况稳定性优化提供了有效解决方案。更多还原

【Abstract】 To address the instability of distributed drive electric vehicles under extreme operating conditions(e.g. low adhesion, high-speed steering maneuvers), this paper proposes a hierarchical collaborative control strategy based on K-means++ algorithm partitioning of vehicle state regions. First, an offline training dataset is built based on the CarSim vehicle model, extracting nine-dimensional vehicle stability feature parameter(e.g. yaw rate and sideslip angle). The K-means++ algorithm is utilized to classify the vehicle state into three regions: stable, coordinated, and control domains, with a dynamic weight coordination module designed accordingly. In the upper controller, the discrete sliding mode control algorithm, integrated with particle swarm optimization for tuning reaching law coefficients, generates target additional yaw moments to track ideal yaw dynamics. Comparative experiments with the integral sliding mode algorithm show the superiority of the discrete sliding mode controller in suppressing peak errors and enhancing tracking accuracy. In the lower controller, a quadratic programming model is built based on stability margin constraints to optimize torque distribution among the four wheels, ensuring that the resultant vectors of longitudinal and lateral forces remain within the friction ellipse. CarSim/Simulink co-simulink demonstrates that, under medium-speed low-adhesion conditions, the proposed strategy reduces peak errors in yaw rate and sideslip angle by 77.2% and 11.6% and improves tracking accuracy by 63.13% and 15.19% compared to the integral sliding mode algorithm. Overall, the results indicate the K-means++ state partitioning and hierarchical discrete sliding mode control enhances lateral stability and robustness.更多还原