摘要
粒子滤波(Particle Filter)是一种基于蒙特卡罗(Monte Carlo)的贝叶斯滤波方法,通常的SIR方法存在严重的粒子匮乏现象。用大权值粒子和小权值粒子的组合来取代小权值粒子,可以减小粒子权值方差,增加粒子多样性。仿真结果表明,在状态估计的初期,使得粒子迅速靠近高似然区域,精度得到了大幅度的提高。同时,算法的实时性也有很好的改善,适用于观测噪声和状态噪声较小的情况。
Particle Filter is a Monte Carlo based Bayesian method, and the problem of the SIR particle filter is particle degeneracy. The combination of two particles was used to replace the particle with less weight, which could reduce the variance of weights and improve the diversity of particles. Results of simulation shows that particles can be moved to the high likelihood area quickly, and the precision of estimation is improved greatly at the beginning of state estimation. And the real-time is also improved to a large extent, and the novel algorithm works well in the environments with low noise.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2012年第7期1470-1473,共4页
Journal of System Simulation
关键词
粒子滤波
预处理
权值方差
粒子多样性
实时性
particle filter
pre-processing
variance of weights
diversity of particles
real-time