基于随机化Halton序列的粒子滤波算法研究

Research of particle filter algorithm based on randomized Halton sequences

为了克服传统粒子滤波蒙特卡洛(MC)随机采样粒子之间的间隙过大与层叠,及其产生的采样效率和滤波精度较低的问题,提出一种基于Halton序列的拟蒙特卡洛(QMC)采样粒子滤波算法。该算法在对Halton序列进行随机化、较好地消除其各维之间相关性的基础上,将之应用于粒子采样过程,以代替蒙特卡洛随机采样,得到用均匀分布粒子近似的后验状态概率密度。仿真证实,算法性能要优于传统粒子滤波算法,改善了采样效率与计算精度,且能克服粒子的退化现象。

For conquering the possible large gaps and clusters which arose from Monte Carlo （ MC ） random sampling in tradi- tional particle filter and resulted in low sampling efficiency and accuracy, this paper proposed a particle filtering algorithm, which introduced Halton sequences based quasi-Monte Carlo （QMC）sampling. Firstly,randomized the Hahon sequences, that could break the correlation of the original ones. Applied the randomized sequences to the sampling process to replace the Monte Carlo random sampling, and could get the posterior state probability density represented by the uniformly distributed particles. Simulations show that the particle filtering algorithm is superior to the traditional one, and can improve the sampling efficiency and accuracy. Especially the algorithm can overcome the degradation of particles.