一种确定性的粒子滤波算法

基于序贯蒙特卡罗的粒子滤波已成为研究非线性非高斯估计问题的一个重要途径.该文将统计学中的数论方法应用于粒子滤波,用代表点来代替随机样本,由于随机分布的代表点较蒙特卡罗方法具有更好的收敛速度,对随机分布有更好的代表性,因此可望提高粒子滤波的性能.在简单介绍了粒子滤波及随机分布代表点理论的基础上,将随机分布的F-偏差代表点应用于粒子滤波,对粒子滤波中关键的初始粒子生成、重点密度采样及再采样过程给出了相应的代表点算法,得到了一个包含最少随机操作的确定性粒子滤波算法.仿真结果表明,确定性粒子滤波算法在滤波性能及计算效率均有不同程度的提高.

A Deterministic Particle Filter Algorithm

Particle filter,which is based on the sequential Monte-Carlo method,has been an important method for the non-linear/non-Gaussian Estimation.In this paper,the number-theoretic method in statistics,which uses the representative points rather than random samples,is applied to particle filter.The rep-points of random distribution,which has better convergence rate and representativeness than the random samples,may be used to improve the performance of particle filter.Followed by an introduction of particle filter and rep-points of random distribution,the F-deviation rep-points are used in particle filter.The corresponding rep-points algorithm,which includes the generation of initial particles,sampling of the importance density and resampling,is given,then a deterministic particle filter Algorithm,with few random operator,can be built.The simulation results show that the performance of the algorithm and computational efficiency can be improved to a certain extent.