Distributed Optimal Linear Fusion Predictors and Filters for Systems With Random Parameter Matrices and Correlated Noises
Shuli Sun
Abstract
A Kalman-like recursive distributed optimal linear fusion predictor (RDOLFP) without feedback in the linear unbiased minimum variance sense is presented for multi-sensor discrete-time linear stochastic systems with random parameter matrices and correlated noises. Local predictions from sensors are sent to a fusion center to fuse with a prior fusion predictor. The proposed RDOLFP without feedback achieves better accuracy than distributed fusion predictors described in the literature that only weight fusion of local predictors, but worse accuracy than a centralized fusion predictor. A RDOLFP with feedback that has the same estimation accuracy as a centralized fusion predictor is also presented. Its optimality is strictly proven. The stability and steady-state properties of the proposed fusion predictors are analyzed. Distributed optimal linear fusion filters with and without feedback, based on the proposed RDOLFPs, are also presented. Two examples demonstrate the effectiveness of the proposed algorithms.