Abstract:For the linear discrete stochastic descriptor systems with multisensor, using me smgu, ar value decomposition, it is transformed into two reduced order subsystems, and a new fusion method of constructing the original state fuser based on the transformed state fuser is presented. Using Kalman filtering method, under the linear minimum variance optimal weighted criterion by matrices, diagonal matrices, and scalars, three optimal weighted fusion reduced order descriptor Kalman estimators are presented respectively. They can handle the fused filtering, smoothing, and prediction problems in a unified framework. They can reduce the computational burden, and can improve the local filtering accuracy. The accuracy relations among three fusers and local estimators are proved. In order to compute the optimal weights, the formula of computing the covariance matrices among local filtering errors is presented. A Monte Carlo simulation example shows its effectiveness.