Sequential fusion filtering based on minimum error entropy criterion

Abstract

According to the minimum error entropy (MEE) criterion in information theory learning (ITL), the fusion filtering problem of non-Gaussian system is studied in this paper. Combined with the advantages of sequential fusion filtering (SFF) in dealing with asynchronous sampling and communication delay, two SFF algorithms are designed under MEE criterion. Firstly, the non-Gaussian multi-sensor system is transformed into a group of non-Gaussian single sensor subsystems. Then, by solving the optimal solution of the cost function corresponding to each subsystem, a set of fixed-point equations relating to the subsystems state is obtained. By using the strategies of global iteration and independent iteration to solve the fixed-point equations, two SFF algorithms based on the MEE criterion are designed, respectively. In addition, the performance including the computational complexity, the correlation between two iteration algorithms and their convergence is analyzed. Finally, simulation results indicate that the proposed SFF methods can effectively deal with the state estimation problem of non-Gaussian multi-sensor system, and can achieve similar fusion filtering accuracy.