Abstract:

In the classical interacting multiple model (IMM) algorithm,because of the Gaussian approximation tothe likelihood function and the confusion of probability density function and probability mass function,the obtainedmode probabilities are only the approximations of probability mass,which is a suboptimal result in the sense ofBayes.In order to solve this problem,a reweighted IMM algorithm is proposed based on the correlation among theestimation errors of mode- conditioned filters and the multi- sensor optimal information fusion criterion.In this algo-rithm,the mode probabilities are updated by calculating the cross- covariance matrix of estimation errors,and thenthe filtering results are fused according to the optimal information fusion theory.Theoretical analysis and simulationresults indicate that the estimation accuracy of the proposed algorithm is significantly improved in comparison withthose of the classical IMM algorithm and other IMM- related algorithms which ignore the error correlation.

Key words: information fusion, correlative estimation errors, optimal information fusion criterion, scalar weight, interacting multiple model algorithm