PARAMETER SET ESTIMATION ALGORITHMS FOR TIME-VARYING SYSTEMS

Parameter set estimation (PSE), a class of system identification schem es which aims at characterizing the uncertainty in the identification experiment, is philosophically different from traditional parameter es timation schemes which seek to identify a single point (model) in the parameter space. The literature has seen a good deal of attention paid to PSE techniques in recent years, primarily because it is projected that they will play a vital role in robust identification for control. An important step in current research along these lines is developmen t of PSE algorithms for systems which are time varying in nature; this is particularly true if the identified model set is to be used in an adaptive setting, such as for gain scheduling or autotuning. In this p aper, we extend an ellipsoid algorithm for PSE of time-invariant syste ms to time-varying systems. We show how knowledge of dependences in th e parameter variations can be exploited to reduce the number of comput ations in the resulting algorithm. Finally, scalar bound inflation, a second strategy for PSE of time-varying systems, is optimized for volu me, and a comparison of the two algorithms is made.