Abstract: In order to accurately obtain vehicle motion state information without relying on the accuracy of the dynamics model, a vehicle state estimation algorithm was proposed based on WOA-SVR. Firstly, by analyzing the basic characteristics of vehicle dynamics, a SVR architecture was designed for estimating the separation of lateral velocity, yaw rate, and vehicle speed. Then, the SVR model was trained on a dataset composed of multiple driving conditions, and the WOA was used to optimize the penalty factor c and kernel function parameter g in the relaxation variables during the training processes. Finally, the estimation algorithm was validated through virtual simulation of single line shift and frequency sweep tests, as well as ABS braking and double line shift actual vehicle tests. The results show that this algorithm effectively improves estimation accuracy and is robust to changes in speed, enabling accurate estimation of vehicle motion states without relying on dynamics models.

Key words: vehicle state estimation, dynamics model, machine learning, support vector regression（SVR）, whale optimization algorithm（WOA)

摘要： 为了不依赖动力学模型精度而准确地获取车辆运动状态信息，提出一种基于鲸鱼优化算法支持向量回归(WOA-SVR)的车辆状态估计算法。首先通过分析车辆动力学基本特性，设计了侧向速度、横摆角速度与车速分离的支持向量回归估计架构；然后对支持向量回归（SVR）模型进行多种行驶工况组成的数据集训练，在训练过程中运用鲸鱼优化算法对松弛变量中的惩罚因子c与核函数参数g进行寻优；最后对估计算法进行单移线、扫频试验虚拟仿真和实车ABS制动、双移线试验验证。结果表明，该算法有效提高了估计精度，且对车速的变化具有鲁棒性，可以实现准确的不依赖动力学模型精度的汽车运动状态估计。

关键词: 车辆状态估计, 动力学模型, 机器学习, 支持向量回归, 鲸鱼优化算法