自适应SDV-UPF算法及其在紧组合中的应用

Adaptive SVD-UPF algorithm and application to tightly-coupled integrated navigation

针对粒子滤波存在重要性密度函数难以选取和系统状态协方差阵可能出现的负定性问题,提出一种新的自适应奇异值分解无迹粒子滤波(ASVD-UPF)算法。该算法采用自适应因子修正动力学模型误差,通过奇异值分解抑制系统状态协方差矩阵的负定性,并以改进的UKF算法产生重要性密度函数,以弥补粒子滤波的缺陷,使该算法适用于非线性、非高斯系统模型的滤波计算。将提出的算法应用到所设计的GPS/SINS/PL紧组合导航系统中进行仿真验证,结果表明,提出算法的经、纬度误差、速度误差和姿态误差范围分别控制在(-0.5″,+0.5″)、(-0.8 m/s,+0.8 m/s)和(-1′,+1′)以内,误差估计的精度和收敛速度明显优于UKF和PF算法,能提高组合导航系统的解算精度。

In view that the importance density function in particle filter is hard to select and the system state covariance matrix is negative definiteness, a new adaptive unscented particle filter based on singular value decomposition （ASVD-UPF） was proposed. The new algorithm is capable of correcting dynamic model error by adaptive factor and restraining the negative definiteness of system state covariance matrix by SVD as well as generating the importance density function to offset defects of PF. With the above advantages, the proposed algorithra is suitable to be applied to the filter calculation of nonlinear and non-Gaussian system model. The proposed algorithm was applied to the GPS/SINS/PL tightly-coupled integrated navigation system and the simulation test was made. The simulation results show that the errors of longitude, latitude, velocity and attitude could be controlled to under （-0.5＂, ＋0.5＂）, （-0.8 m/s, ＋0.8 m/s） and （-1＇, ＋1＇） respectively by the proposed algorithm. It is proved that the error estimation precision and convergence speed of the proposed algorithm are better than that of UKF and PF, and the positioning precision of integrated navigation is improved.