ON ENTRY-WISE ORGANIZED FILTERING

SUZDALEVA, E.

Abstract

The paper deals with Bayesian view on estimation of time-varying parameters and adaptivity. The Bayesian approach is especially productive when the parameter estimation is a part of adaptive control and is taken in a closed control loop. From that point of view the problem of state estimation and output prediction in a factorized form is discussed. The factorized form is helpful for observing the dependence both of the output and next state factors from the state factors. A special transformation of the state leading to a triangular matrix of coefficients is considered. The paper factorizes the state-space model and describes the idea of calculating the time and data update recursion. The idea opens the way to manage the mixed models as the joint - continuous and discrete - probability distributions. The expected result is supposed to be applied to the filtering problem.

Coresponding author e-mail: suzdalev[at]utia[dot]cas[dot]cz

Session: Modelling, Simulation, and Identification of Processes