A MOVING HORIZON-BASED APPROACH FOR LEAST-SQUARES ESTIMATION

A general formulation of the moving horizon estimator is presented. An algorithm with a fired-size estimation window and constraints on stat es, disturbances, and measurement noise is developed and a probabilist ic interpretation is given. The moving horizon formulation requires on ly one more tuning parameter (horizon size) than many well-known appro ximate nonlinear filters such as extended Kalman filter filter (EFK), iterated EKF, Gaussian second-order filter, and statistically lineariz ed filter. The choice of horizon size allows the user to achieve a com promise between the better performance of the batch least-squares solu tion and the reduced computational requirements of the approximate non linear filters. Specific issues relevant to linear and nonlinear syste ms are discussed with comparisons made to the Kalman filter, EKF, and other recursive and optimization-based estimation schemes.