路面附着系数的自适应衰减卡尔曼滤波估计

为了获得实时、准确的路面附着系数，进一步提高观测路面附着系数算法的精度和收敛速度，结合非线性车辆动力学模型和轮胎力修正模型，搭建分布式驱动电动汽车联合仿真平台，提出一种基于自适应衰减无迹卡尔曼滤波的路面附着系数观测算法。该算法设计与各轮对应的路面附着系数观测器，应用协方差匹配判据对观测器发散趋势进行判别，设计自适应加权系数修正预测协方差，以增强新近观测数据的利用率；同时采用次优Sage-Husa噪声估计器对未知的系统过程噪声进行估计，抑制观测器的记忆存储长度，调整过程噪声和测量噪声的均值与协方差，提高观测器的跟踪能力。利用分布式驱动电动汽车分别进行高、低附着路面和对开路面直线制动试验，并将自适应衰减无迹卡尔曼滤波路面附着系数观测器的观测结果与无迹卡尔曼滤波观测值、参考路面附着系数进行比较和分析。结果表明：高附着路面条件下，所设计的算法估计误差可控制在0.64%以内；低附着路面条件下，所设计的算法估计误差可控制在1.03%以内；对开路面条件下估计误差可控制在1.26%以内；自适应衰减无迹卡尔曼滤波算法相比无迹卡尔曼滤波算法响应速率更快，具有更高的估计精度和较强的自适应能力，估计结果整体上维持稳定，能够适应各种不同路面的估计。

Estimation Algorithm for Road Adhesion Coefficient Using Adaptive Fading Unscented Kalman Filter

To obtain real-time and accurate road adhesion coefficient and improve the accuracy and convergence speed of the observation algorithm for road adhesion coefficient, a co-simulation platform for a distributed drive electric vehicle was built based on nonlinear vehicle dynamics and tire force correction models. Thereafter, an adaptive fading unscented Kalman filter algorithm on road adhesion coefficient was proposed. In this algorithm, the road adhesion coefficient observers were designed for each wheel and the covariance matching criterion was subsequently applied to discriminate the divergence trend of the observers. An adaptive weighting coefficient was designed to modify the predicted covariance, thereby enhancing the utilization rate of the latest observed data. Simultaneously, the suboptimal Sage-Husa noise estimator was used to estimate the unknown system process noise, which suppressed the memory storage length of the observer. Subsequently, the mean and covariance of the process and measurement noises were adjusted to improve the tracking ability of the observer. The braking tests on high/low attached roads and bisectional roads were carried out by a distributed drive electric vehicle. The adhesion coefficients obtained by the adaptive fading unscented Kalman observer were analyzed and compared with those obtained by the unscented Kalman filter and the reference values. The results of the proposed algorithm illustrated that the estimation errors were controlled within 0.64%, 1.03%, and 1.26% under the high attached road surface, low attached road surface, and bisectional road surface, respectively. The adaptive fading unscented Kalman filter algorithm had a faster convergence speed, higher estimation accuracy, and stronger adaptive ability than the unscented Kalman filter algorithm. The estimation results of the proposed algorithm are stable and adaptable to the real-time estimation of different road surfaces.