アンサンブル平均を用いた粒子フィルタ

アンサンブルの統合方法に関する検討

抄録

This paper proposes a novel particle filter using ensemble average. The proposed method is constructed by some particle sets (ensambles) that have different initial hyper-parameters, and performs same operation of a particle filter algorithm to each ensamble. The proposed method can estimate both system states and hyper parameters of the system, by weighted average that is calculated by the logarithm likelihood of each ensemble and the spread of ensembles. In this paper, to compare the conventional particle filter with the novel particle filters, we tests the proposed method under the one dimension nonlinear system and observation model.