Estimation of Large-Scale Implicit Models Using 2-Stage Methods
Rolf Henriksen · Modeling Identification and Control A Norwegian Research Bulletin · 1985
The problem of estimating large scale implicit (non-recursive) models by two- stage methods is considered. The first stage of the methods is used to construct or estimate an explicit form of the total model, by constructing a minimal stochastic realization of the system. This model is then subsequently used in the second stage to generate instrumental variables for the purpose of estimating each sub-model separately. This latter stage can be carried out by utilizing a generalized least squares method, but most emphasis is put on utilizing decentralized filtering algorithms and a prediction error formulation. A note about the connection between the original TSLS-method (two-stage least squares method) and stochastic realization is also made.