基于新息和残差的自适应UKF算法

An Innovation and Residual-Based Adaptive UKF Algorithm

针对先验噪声统计特性与实际不符引起卡尔曼滤波精度下降的情况,提出了一种基于新息和残差序列在线估计噪声统计特性的自适应无迹卡尔曼滤波算法。该算法首先通过新息序列实现了对观测噪声协方差矩阵的实时跟踪;然后根据新息和残差的正交性原理估计过程噪声协方差的实时变化;最后利用协方差匹配原则在线修正噪声的理论协方差使其逼近真实的噪声水平,从而实现最优估计。另外算法中通过引入尺度因子,进一步减小了泰勒展开造成的截断误差,提高了估计的精度。DR/GPS组合导航系统的仿真实验表明,该算法对时变的噪声统计特性有较强的自适应性,滤波精度更高,鲁棒性更强。

Considering that the prior statistics noise of a Kalman filter does not agree with its real behavior,an adaptive unscented Kalman filter algorithm based on innovation and residual sequences is proposed to on-line estimate the statistics noise property. First,the observation noise covariance matrix is real-time tracked through innovation sequence. Then the real time variation of process noise covariance is on line estimated based on the orthogonal principle between innovated sequence and residual sequence estimated. Finally,the theoretical covariance is on line adjusted to real noise level based on the covariance matching principle,thus achieving the optimum estimation. Additionally,a scale factor is introduced to reduce the truncation error caused by the Taylor series expansion,thus improving estimation accuracy. Simulations are performed in a DR/GPS integrated navigation system. The results show that the proposed algorithm is more adaptive to the time-varied noise statistics; the filtering accuricy and robustness are also improved.