• PDF
• Cite
• Share
Article Contents  Article Contents

# Resampled ensemble Kalman inversion for Bayesian parameter estimation with sequential data

• * Corresponding author: Jinglai Li

The work was supported by NSFC under grant number 11771289

• Many real-world problems require to estimate parameters of interest in a Bayesian framework from data that are collected sequentially in time. Conventional methods to sample the posterior distributions, such as Markov Chain Monte Carlo methods can not efficiently deal with such problems as they do not take advantage of the sequential structure. To this end, the Ensemble Kalman inversion (EnKI), which updates the particles whenever a new collection of data arrive, becomes a popular tool to solve this type of problems. In this work we present a method to improve the performance of EnKI, which removes some particles that significantly deviate from the posterior distribution via a resampling procedure. Specifically we adopt an idea developed in the sequential Monte Carlo sampler, and simplify it to compute an approximate weight function. Finally we use the computed weights to identify and remove those particles seriously deviating from the target distribution. With numerical examples, we demonstrate that, without requiring any additional evaluations of the forward model, the proposed method can improve the performance of standard EnKI in certain class of problems.

 Citation: • • Figure 1.  The simulated data for $\sigma = 0.4$ (left) and $\sigma = 0.6$ (right). The lines show the simulated states in continuous time and the dots are the noisy observations

Figure 2.  The results for the case where noise variance is $0.4^2$. Left: the error for the simulation with $100$ particles. Right: the error for simulation with $500$ particles

Figure 3.  The results for the case where noise variance is $0.6^2$. Left: the error for the simulation with $100$ particles. Right: the error for simulation with $500$ particles

Figure 4.  $\Delta = 0.05.$ Top: the average estimation error of the simulation with 2000 particles. Bottom: the average estimation error of the simulation with 5000 particles. In both rows, the left figure shows the results of the observed dimensions and the right one shows the unobserved ones

Figure 5.  $\Delta = 0.1.$ Top: the average estimation error of the simulation with 2000 particles. Bottom: the average estimation error of the simulation with 5000 particles. In both rows, the left figure shows the results of the observed dimensions and the right one shows the unobserved ones

Figure 6.  Left: The ground truth for x. Right: Both the noise-free and the noisy data at t = 3

Figure 7.  The estimation error for the high dimensional nonlinear example. Left: the results for $M = 2000$; Right: the results for $M = 5000$

• ## Article Metrics  DownLoad:  Full-Size Img  PowerPoint