一种高精度自适应粒子滤波算法

Investigation of a high-precision algorithm for adaptive particle filtering

针对非线性非高斯系统的状态估计问题,提出一种新的高精度自适应粒子滤波算法.该算法采用有限差分扩展卡尔曼滤波器产生优选的建议分布函数,融入最新量测信息,有效克服了粒子退化问题;考虑到预测误差对粒子采样效率的影响,引入系统估计和预测提供的新息差值,通过新息差值在线自适应调整采样粒子数,较好地保证了粒子采样的高效性.理论分析和实验结果表明改进的滤波算法具有较高的滤波精度,是一种高精度自适应粒子滤波算法.

A new improved particle filtering algorithm was proposed for state estimation of non-Gaussian nonlinear systems.In this algorithm finite-difference extended Kalman filter was adopted to produce an optimized proposal distribution function and the updated measurement messages could be merged in,so that the particle degradation would be avoided effectively.Taking account of the influence of predicted errors on particle sampling efficiency,a deviation value of updated message between system estimation and prediction was introduced.Then,by means of on-line adaptive adjustments of sampling particle count with this deviation value,a high efficiency of particle sampling was ensured.Theoretical analysis and experimental results showed that the improved filtering algorithm exhibited higher filtering precision and would be a kind of high-precision adaptive particle filtering algorithm.