Ensemble Kalman Filter using Gaussian-Sum Predicted State Probability Density Functions

抄録

We propose to use Gaussian-sum predicted state probability density functions (PDFs) in the algorithm of the ensemble Kalman lter (EnKF) to enhance its l tering accuracy. We analyze the EnKF in terms of the moment-matched linearization for the nonlinear observation model and show that the ltering accuracy of the EnKF can be improved by using the Gaussian-sum predicted state PDFs. We numerically con rm the effectiveness of the new lters through simulations using benchmark ltering problems of the vector nonlinear growth model and the satellite reentry.