Reduced Order Estimation

Abstract

When dealing with large scale systems, sometimes it is not necessary to estimate the complete state vector. Rather, one might be interested in only some state variables or a linear combination of the state vector which is of smaller dimension than the original system. In this case it is not economical, and maybe, not feasible to design a full order Kalman filter. It is more attractive from at least computational and economical reasons to design a reduced order filter. The objective here is to design such reduced-order filters to estimate a set of desired variables. This problem was addressed by many investigators. For -example, in (1] the authors derived an unbiased filter provided that the desired and the measurable variables satisfy some rank conditions. The procedure presented here is based on an appropriate Ressenberg [21 representation. The desired variables are viewed as the states of a subsystem driven by the interface variables. Additional measurements on these interface variables are required to obtain an unbiased filter. Conditions for the stability of the filter are derived6