基于极大似然准则的INS/GNSS组合导航自适应UKF滤波算法

Maximum likelihood principle based adaptive unscented Kalman filter for INS/GNSS integration

为提高INS/GNSS组合系统对过程噪声方差不确定性的鲁棒性,提出一种基于极大似然准则的自适应UKF算法。在该算法中,首先利用新息向量的统计信息构造量测向量的后验概率密度,然后通过极大似然准则在线求取过程噪声方差的估值,并将其反馈至UKF滤波过程,用于调整卡尔曼增益矩阵。提出的算法可以抑制过程噪声方差不确定性对滤波解的影响,克服了UKF的缺陷。仿真结果表明,当过程噪声的标准差增大为其真实值的4倍时,相比于UKF,提出方法的导航精度可至少提高45.5%;相比于ARUKF,其导航精度也可至少提高35.7%。跑车实验结果也验证了提出算法的有效性。

A maximum-likelihood based adaptive unscented Kalman filter is presented to improve the robustness of INS/GNSS integrated system against uncertainty of process noise. Firstly, the posterior probability density of the measurement vector is constructed by using the statistics of the innovation vector. Subsequently, the estimation of process noise covariance is obtained based on the maximum-likelihood principle, and are then fed back to the UKF filtering procedure to adjust Kalman gain matrix. The proposed filter has the capability to resist the impact of the process noise covariance uncertainty on filtering solution, overcoming the limitation of the UKF. Simulation results demonstrate that, the navigation accuracy achieved by the proposed method is at least 45.5% higher than that by UKF and 35.7% higher than that by ARUKF, as the standard deviation of process noise is enlarged to four times of its true value. The effectiveness of the proposed filter is also verified via running-car experiment results.