Distributed Cubature Kalman Filter based on MEEF with Adaptive Cauchy Kernel for State Estimation

TL;DR

This paper introduces a novel distributed cubature Kalman filter that uses adaptive minimum error entropy with Cauchy kernels to improve robustness against non-Gaussian noise and abnormal data, while reducing communication load in sensor networks.

Contribution

It proposes the AMEEF-DCKF algorithm combining adaptive kernel bandwidth optimization and a leader-follower consensus method for enhanced state estimation in multi-sensor networks.

Findings

Improved robustness against non-Gaussian noise and abnormal data.

Reduced communication burden through leader-follower consensus.

Validated effectiveness in power system and vehicle navigation scenarios.

Abstract

Nowadays, with the development of multi-sensor networks, the distributed cubature Kalman filter is one of the well-known existing schemes for state estimation, for which the influence of the non-Gaussian noise, abnormal data, and communication burden are urgent challenges. In this paper, a distributed cubature Kalman filter based on adaptive minimum error entropy with fiducial points (AMEEF) criterion (AMEEF-DCKF) is proposed to overcome the above limitations. Specifically, firstly, in order to solve the influence of various types of non-Gaussian noise and abnormal data, the AMEEF optimization criterion is designed, in which the kernels used are Cauchy kernels with adaptive bandwidth. At the same time, the designed optimization criterion has enhanced the numerical stability and optimized the kernel bandwidth value. Next, in order to address the communication burden problem in multi-sensor networks, where a leader and a follower are distinguished, a distributed algorithm is constructed to achieve an average consensus among these sensors, called leader-follower average consensus (LFAC). Additionally, the convergence proof of the average consensus algorithm and the computational complexity analysis of the AMEEF-DCKF algorithm are also presented. Finally, through a 10-node sensor network, the effectiveness of the proposed algorithm is demonstrated in estimating the state of the power system and navigating land vehicles in complex environments.