Based on the decomposition in canonical form, an optimal state filter in the linear unbiased minimum variance sense is given for single-sensor stochastic singular systems with unknown disturbance and correlated noises in the case of Y-observable system, which is independent of the unknown disturbance. When the system is measured by multiple sensors, the computation formula for the filtering error cross-covariance matrix between any two sensor subsystems is derived. Further, the distributed information fusion state filter is given based on the fusion algorithm weighted by matrix in the linear minimum variance sense. The simulation research shows the effectiveness.