权值自适应调整Unscented粒子滤波及其在组合导航中的应用

Unscented particle filtering with adaptive adjusted weight and its application in integrated navigation

针对粒子滤波存在的重要性密度函数难以选取和粒子退化问题,提出了一种新的权值自适应调整Unscented粒子滤波算法。该算法在Unscented粒子滤波的采样过程中吸收权值自适应调整的优点,考虑最新量测影响,通过欧氏距离和反映量测噪声统计特性的精度因子来自适应的调整粒子对应权值分布,增加有用粒子的权值,降低粒子退化程度,保持粒子多样性。同时Unscented变换提高了滤波精度,使该算法能更好地适用于非线性、非高斯系统模型的计算。将提出的算法应用于GPS/DR组合导航系统进行仿真验证,结果表明,提出的权值自适应调整Unscented粒子滤波算法得到的东向定位误差控制在±5.5 m附近,北向定位误差则在±5.2 m附近,滤波性能明显优于扩展卡尔曼滤波和Unscented粒子滤波,能提高GPS/DR组合导航系统解算精度。

Particle filtering causes degeneration and has difficulties in selecting the importance density function. Aiming at this problem, this paper proposes a weight adaptive adjustment unscented particle filtering algorithm, which adds the concept of weight adjustment to the sample process of unscented particle filtering. This algorithm can adaptively adjust the weight function according to the latest measurement, the Euclidean distance and the accuracy factor constructed from statistic performance of measurement information, thus increasing the efficient weights and prevent particle from degeneracy and maintain particle diversity. It also uses the unscented transformation to improve the accuracy of particle filtering, therefore it is more suitable for the filtering calculation of a nonlinear and non-Gaussian model. The proposed algorithm has been applied to GPS/DR integrated navigation system. Experiments and comparisons demonstrate that the east and north position error of weight adaptive adjustment unscented particle filtering are within ＋ 5.5 m and ＋ 5.2m respectively. This algorithm is better than the extended Kalman Filtering and unscented particle filtering algorithms in terms of accuracy, and also improves the calculation precision in GPS/DR integrated navigation system.