摘要
针对辅助变量粒子滤波(AVPF)对状态估计无法获得较好的滤波精度的问题,本文提出了一种改进的辅助变量粒子滤波算法.将正则化的思想引入到辅助变量粒子滤波的重采样中,在重采样中将离散的概率分布函数近似为连续的分布函数,该方法不仅保留了AVPF在重采样之前依据似然值的大小对原粒子集中的各个权值进行修正,从经过平滑后的后验密度中重采样的特点,在重采样中引入正则化思想后还能够保持粒子的多样性,增加有效样本数目,能够有效抑制样本退化.针对一个被广泛采用的双峰,高度非线性的系统模型,在选取不同的过程噪声下,进行Monte Carlo仿真实验.仿真实验表明,改进的辅助变量粒子滤波具有更好滤波精度.
Auxiliary variable particle filter （AVPF） at smaller process noise, state estimation can not get better filtering accuracy, this paper presents a novel auxiliary variable particle filter. Regularization will be introduced into resampling of auxiliary variable particle filter, In the resampling, the discrete probability distribution function approximated by a continuous distribution function, which not only retains the feature that AVPF based on the size of likelihood values revises the original particles are concentrated weights before resampling , from smooth posterior density resarnpling, after the introduction of regularization also able to maintain the diversity of particle in the resampling, increasing the number of the effective particles, can effectively inhibit the degradation of the sample For doublet, a highly nonlinear system model has been widely used , in the process of selecting a different noise, carry out Monte Carlo simulations. Simulation results show that the novel variable particle filter has better filtering accuracy.
出处
《微电子学与计算机》
CSCD
北大核心
2017年第2期39-42,共4页
Microelectronics & Computer
关键词
辅助变量粒子滤波
正则化
概率分布函数
auxiliary variable particle filter
regularization
probability distribution function
