针对多传感器系统动态偏差估计问题,在不敏粒子滤波(UPF)算法的基础上,提出了一种修正的不敏粒子滤波(Modi-fied UPF,MUPF)算法.由于系统动态偏差引起的异常量测值时,MUPF算法利用滤波预测残差构建的调节因子控制新息协方差矩阵,进而调整滤波增益的大小；在不丢失有用新息的前提下,减小了异常量测对滤波估计结果的影响.利用上述算法与不敏卡尔曼滤波(UKF)算法和扩展卡尔曼粒子滤波(EPF)算法进行了仿真比较.结果表明,MUPF算法对系统动态距离和角度偏差估计的均方根误差(RMSE)明显小于UKF算法和EPF算法的估计结果,提高了估计精度和可靠性.显然,MUPF算法也适用于系统固定测量偏差估计和目标状态估计.%Focus on the dynamic errors estimation problem of multi - sensor systems, a Modified UPF (Unscented Particle Filter) algorithm was provided based on the UPF algorithm. It uses a scaling factor formed by filtering prediction residual to control the innovation covariance matrix and then to adjust the filter gain when some abnormal measurements exist due to system dynamic errors. It reduces the impact of the abnormal measurement on filter estimates without loss of useful innovations. The advanced technique was compared with the UKF (Unscented Kalman filter) algorithm and EPF (Extended Kalman Particle Filter) algorithm by computer simulation. The results justify that the Root - mean - square Error (RMSE) of MUPF algorithm about system dynamic distance and angle biases estimations is obviously lower than the corresponding of UKF and EPF estimations. Thus it improves the biases estimation accuracy and reliability. Apparently, the MUPF algorithm can also be applied to fixed measurement bias estimation and target state estimation.
展开▼
