Consensus cubature filtering based on Gaussian process for distributed sensor network with model uncertainty

2022 
Abstract In this paper, we consider a distributed sensor network without accurate knowledge of the state and observation models or even the form of system functions. Traditionally, consensus filtering approaches are widely used for distributed state estimation, but they require an accurate parametric model of the system. We propose a novel consensus cubature filtering algorithm based on Gaussian process (CCF-GP) for distributed state estimation with model uncertainty. The parameters in distributed sensor networks can be obtained from training data by Gaussian process regression (GPR). The approximate parametric model can be incorporated into Gaussian process (GP) to enhance the distributed estimation performance. In the Bayesian filtering frameworks, we derive the prediction and update steps of the hybrid consensus on measurements and consensus on information (HCMCI) method based on GPR. And then every node in the distributed network cooperates the information pair with its neighbors. For nonlinear systems, we derive the GP-based consensus filtering based on the spherical-radial cubature rule. Comparing with the standard nonlinear consensus filtering methods, the validity of our proposed algorithm is confirmed by a series of simulations.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    38
    References
    0
    Citations
    NaN
    KQI
    []