CAJ | 학술논문

传统粒子滤波(PF)中,重采样步骤里存在着粒子的"平均化"现象,导致粒子本身概率大小的因素被忽略,没有充分利用粒子集所包含的信息.通过改进抛弃小权值粒子的原则,以及充分利用粒子权值大小所代表的意义来进行粒子复制的两点进行算法改进,采用一维非线性目标跟踪模型和新的二维动态跟踪模型分别研究改进PF算法对于平均RMSE的影响.通过仿真,证明了改进后的算法可以显著降低变量的平均RMSE,特别是在二位动态跟踪模型中,使位置坐标和速度两种变量的平均均方根误差(RMSE)都有所改善,从而提高了滤波性能.
전통입자려파(PF)중,중채양보취리존재착입자적"평균화"현상,도치입자본신개솔대소적인소피홀략,몰유충분이용입자집소포함적신식.통과개진포기소권치입자적원칙,이급충분이용입자권치대소소대표적의의래진행입자복제적량점진행산법개진,채용일유비선성목표근종모형화신적이유동태근종모형분별연구개진PF산법대우평균RMSE적영향.통과방진,증명료개진후적산법가이현저강저변량적평균RMSE,특별시재이위동태근종모형중,사위치좌표화속도량충변량적평균균방근오차(RMSE)도유소개선,종이제고료려파성능.
In standard Particle Filtering (PF) algorithm, the "average problem" of particles replication in the re-sampling step neglects the significance of the probability of particles and does not sufficiently use the information con-tained in the particle pool. This paper improves the algorithm via reasonably setting the principle of abandoning small weight particles and replicating particles by the knowledge of weights. Experiments of one dimensional target tracing model and x-y target tracing model validated that the improved PF can ameliorate the average RMSE of variables. Especially in x-y target tracing model, better outcome can be got both in position and velocity. It is concluded that the improved algorithm can markedly reduce the RMSE of variables, and therefore the filtering performance of im-proved PF is greatly superior to that of the standard PF.