A regularized kernel-based method for learning a module in a dynamic network with correlated noise

Venkatakrishnan C. Rajagopal, Karthik R. Ramaswamy, Paul M.J. Van Den Hof

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

Samenvatting

In this paper, we consider the problem of identifying one system (module) embedded in a dynamic network that is disturbed by colored process noise sources, which can possibly be correlated. To achieve this using the direct method for single module identification, we need to formulate a Multi-Input-Multi-Output (MIMO) estimation problem which requires model order selection step for each module in the setup and estimation of large number of parameters. This results in a larger variance in the estimates and an increase in computation complexity. Therefore, we extend the Empirical Bayes Direct Method [1], which handles the above mentioned problems for a Multi-Input-Single-Output (MISO) setup to a MIMO setting by suitably modifying the framework. We keep a parametric model for the desired target module and model the impulse response of all the other modules as independent zero mean Gaussian process governed by a first-order stable spline kernel. The parameters of the target module are obtained by maximizing the marginal likelihood of the output using the Empirical Bayes (EB) approach. To solve this, we use the Expectation Maximization (EM) algorithm which offers computational advantages. Numerical simulation illustrate the advantages of the developed method over existing classical methods.

Originele taal-2Engels
Titel2020 59th IEEE Conference on Decision and Control, CDC 2020
UitgeverijInstitute of Electrical and Electronics Engineers
Pagina's4348-4353
Aantal pagina's6
ISBN van elektronische versie978-1-7281-7447-1
DOI's
StatusGepubliceerd - 14 dec 2020
Evenement59th IEEE Conference on Decision and Control, CDC 2020 - Virtual, Jeju Island, Zuid-Korea
Duur: 14 dec 202018 dec 2020

Congres

Congres59th IEEE Conference on Decision and Control, CDC 2020
LandZuid-Korea
StadVirtual, Jeju Island
Periode14/12/2018/12/20

Vingerafdruk Duik in de onderzoeksthema's van 'A regularized kernel-based method for learning a module in a dynamic network with correlated noise'. Samen vormen ze een unieke vingerafdruk.

Citeer dit