.png){kind=icon}
.png){kind=icon}
Return
A Bayesian filtering network for state estimation with unknown system dynamics
.png){kind=icon}
PRE
AI
DOI：10.1016/j.sigpro.2025.110365.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To address the challenge of Bayesian filtering with unknown state transition model, we propose a Bayesian optimization framework using stochastic variational inference (BOSVI) for accurate modeling and state estimation. Specifically, the proposed BOSVI framework consists of a prior prediction network and a posterior correction network. First, the prior network models the latent system dynamics using a parameterized stochastic differential equation (SDE), allowing flexible approximation of nonlinear state evolution function. Then, an evidence lower bound (ELBO) is derived to optimize the SDE parameters and yield a prior distribution over the system states. Meanwhile, to correct deviations in the prior estimates, we design a neural network using real-time observations for posterior updates, which integrates GRU and self-attention mechanisms to dynamically refine state estimates and reduce uncertainty. Finally, simulation results demonstrate that, compared to other representative algorithms, the proposed BOSVI achieves superior estimation performance under various perturbation environments and observation mismatches.
Journal
IF:
3.6
Papers:
9.9K
Citations:
1.7W