Filtering, Prediction and Smoothing with Point-Mass Approach

Abstract

The point-mass method for nonlinear state estimation is re-examined. Several aspects of the method are treated and algorithms for multi-step prediction and smoothing based on the point-mass filtering algorithm are derived. Thus solution of all three estimation problems, i.e. filtering, prediction and smoothing, is unified within the pointmass framework. An anticipative technique for adaptation of support grid is presented. This technique automatically sets the number of grid points according to a future behaviour of the system and with respect to the user-defmed accuracy of pdf approximation.