LINEAR FILTRATION OF RANDOM SEQUENCES USING A LEAST SQUARE METHOD WITH REGULARIZATION

Abstract

At high problem dimension the Kalman filter becomes difficult to realize in real time due to the high computational costs. Alternatively, a technique of filter synthesis on the basis of the extended least square method with extended square discrepancy is given. The technique allows to reduce the computation costs for the search of filter gain coefficients, but it increases the variance of the filtering error in comparison with Kalman filter. The degree of this increase in case of prior information is not taken into account is shown on example.