𝔖 Bobbio Scriptorium
✦   LIBER   ✦

On the problem of ambiguities in maximum likelihood identification

✍ Scribed by T. Bohlin

Publisher
Elsevier Science
Year
1971
Tongue
English
Weight
1015 KB
Volume
7
Category
Article
ISSN
0005-1098

No coin nor oath required. For personal study only.

✦ Synopsis

When modelling a physical process, basic mathematical assumptions are often difficult to verify in advance. However, the applicability of certain statistical identification methods to a given data sample can be checked by testing the resulting model.

Summary--This contribution derives new large-sample properties of the result of Maximum-Likelihood identification of a discrete-time process, characterized by an unknown, rational input-transfer function G and an unknown, rational noise-transfer function F. The purpose is to develop some means to detect possible false results of such an identification, caused by unknowingly violating prerequisites for the identification, e.g. by incorrect model structure or order, or having reached a local maximum in the search. Also, the effect of linear feedback and the identifiability in dosed loop are discussed.

Results are applied to artificially generated data and to plant data---collected in closed loop from the drying process in paper making.

πŸ“œ SIMILAR VOLUMES

On the uniqueness of maximum likelihood
On the uniqueness of maximum likelihood identification
✍ T SΓΆderstrΓΆm πŸ“‚ Article πŸ“… 1975 πŸ› Elsevier Science 🌐 English βš– 426 KB

The maximum likelihood method of identification is a powerful tool for obtaining mathematical models of dynamic processes. To apply this method a loss function has to be minimized. The aim of the paper is an investigation of the local minimum points of this loss function for a common structure of a

Comments on β€˜non-convergence of the appr
Comments on β€˜non-convergence of the approximate maximum likelihood identification algorithm’
✍ Lennart Ljung πŸ“‚ Article πŸ“… 1980 πŸ› Elsevier Science 🌐 English βš– 130 KB

The following three comments and claims are made: 1. The 'approximate maximum likelihood method ' [called RELS in (S6derstr6m et al., 1978)] may work well in applications even though it has been proven that it does not always converge. 2. It was incorrect in (\* Ljung et al., 1975) to call the 'si

Non-convergence of the approximate maxim
Non-convergence of the approximate maximum likelihood identification algorithm
✍ V. Panuska πŸ“‚ Article πŸ“… 1980 πŸ› Elsevier Science 🌐 English βš– 205 KB

The question of non-convergence of the approximate maximum likelihood identification algorithm is discussed. It is pointed out that although there are systems for which in theory the algorithm does not converge to any finite limit, the computer implementation always gives results which settle at con