This paper addresses the computational problem of fixed-interval smoothing state estimation in linear time-varying Gaussian stochastic systems.A new fixed-interval Kalman smoothing algorithm is proposed,and the corres...This paper addresses the computational problem of fixed-interval smoothing state estimation in linear time-varying Gaussian stochastic systems.A new fixed-interval Kalman smoothing algorithm is proposed,and the corresponding form of the smoother is derived.The method is able to accommodate situations where process and measurement noises are correlated,a limitation often encountered in conventional approaches.The Kalman smoothing problem discussed in this paper can be reformulated as an equivalent constrained optimization problem,where the solution corresponds to a set of linear equations defined by a specific co-efficient matrix.Through multiple permutations,the co-efficient matrix of linear equations is transformed into a block tridiagonal form,and then both sides of the linear system are multiplied by the inverse of the co-efficient matrix.This approach is based on the transformation of linear systems described in the SPIKE algorithm and is particularly well-suited for large-scale sparse block tridiagonal matrix structures.It enables efficient,parallel,and flexible solutions while maintaining a certain degree of block diagonal dominance.Compared to directly solving block tridiagonal co-efficient matrices,this method demonstrates appreciable advantages in terms of numerical stability and computational efficiency.Consequently,the new smoothing algorithm yields a new smoother that features fewer constraints and broader applicability than traditional methods.The estimates,such as smoothed state,covariance,and cross-covariance,are essential for fields,such as system identification,navigation,guidance,and control.Finally,the effectiveness of the proposed smoothing algorithm and smoother is validated through numerical simulations.展开更多
文摘This paper addresses the computational problem of fixed-interval smoothing state estimation in linear time-varying Gaussian stochastic systems.A new fixed-interval Kalman smoothing algorithm is proposed,and the corresponding form of the smoother is derived.The method is able to accommodate situations where process and measurement noises are correlated,a limitation often encountered in conventional approaches.The Kalman smoothing problem discussed in this paper can be reformulated as an equivalent constrained optimization problem,where the solution corresponds to a set of linear equations defined by a specific co-efficient matrix.Through multiple permutations,the co-efficient matrix of linear equations is transformed into a block tridiagonal form,and then both sides of the linear system are multiplied by the inverse of the co-efficient matrix.This approach is based on the transformation of linear systems described in the SPIKE algorithm and is particularly well-suited for large-scale sparse block tridiagonal matrix structures.It enables efficient,parallel,and flexible solutions while maintaining a certain degree of block diagonal dominance.Compared to directly solving block tridiagonal co-efficient matrices,this method demonstrates appreciable advantages in terms of numerical stability and computational efficiency.Consequently,the new smoothing algorithm yields a new smoother that features fewer constraints and broader applicability than traditional methods.The estimates,such as smoothed state,covariance,and cross-covariance,are essential for fields,such as system identification,navigation,guidance,and control.Finally,the effectiveness of the proposed smoothing algorithm and smoother is validated through numerical simulations.
