Aarhus University Seal / Aarhus Universitets segl

Ahmad Madary

A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems

Publikation: Bidrag til tidsskrift/Konferencebidrag i tidsskrift /Bidrag til avisKonferenceartikelForskningpeer review

Standard

A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems. / Madary, Ahmad; Momeni, Hamid Reza; Abate, Alessandro; Larsen, Kim Guldstrand.

I: IFAC-PapersOnLine, Bind 54, Nr. 5, 07.2021, s. 259-264.

Publikation: Bidrag til tidsskrift/Konferencebidrag i tidsskrift /Bidrag til avisKonferenceartikelForskningpeer review

Harvard

Madary, A, Momeni, HR, Abate, A & Larsen, KG 2021, 'A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems', IFAC-PapersOnLine, bind 54, nr. 5, s. 259-264. https://doi.org/10.1016/j.ifacol.2021.08.508

APA

Madary, A., Momeni, H. R., Abate, A., & Larsen, K. G. (2021). A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems. IFAC-PapersOnLine, 54(5), 259-264. https://doi.org/10.1016/j.ifacol.2021.08.508

CBE

MLA

Vancouver

Author

Madary, Ahmad ; Momeni, Hamid Reza ; Abate, Alessandro ; Larsen, Kim Guldstrand. / A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems. I: IFAC-PapersOnLine. 2021 ; Bind 54, Nr. 5. s. 259-264.

Bibtex

@inproceedings{7fa79703d05f49608d6f3cf17d9d5bd8,
title = "A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems",
abstract = "In this paper, a two-level Bayesian framework is proposedfor the identification of nonlinear hybrid systems fromlarge data sets by embedding it in a four-stage procedure.At the first stage, feature vector selection techniques areused to generate a reduced-size set from the given trainingdata set. The resulting data set then is used to identifythe hybrid system using a Bayesian method, where theobjective is to assign each data point to a correspondingsub-mode of the hybrid model. At the next (third) stage,this data assignment is used to train a Bayesian classifierto separate the original data set and determine thecorresponding sub-mode for all the original data points.Finally, once every data point is assigned to a sub-mode, aBayesian estimator is used to estimate a regressor for eachsub-system independently. This Bayesian approach finds acompromise between model complexity and accuracy. Theproposed method tested on three case studies. First, theidentification based on the full data set is compared withthat based on the reduced data set, over a nonlinear hybridsystem. Then, the proposed approach is compared with anexisting method from the literature, with improvedperformance over estimation error and mode assignment.Finally, the performance of the new method is tested byidentifying a complex hybrid model. ",
author = "Ahmad Madary and Momeni, {Hamid Reza} and Alessandro Abate and Larsen, {Kim Guldstrand}",
year = "2021",
month = jul,
doi = "10.1016/j.ifacol.2021.08.508",
language = "English",
volume = "54",
pages = "259--264",
journal = "IFAC-PapersOnLine",
issn = "2405-8963",
publisher = "Elsevier",
number = "5",
note = "7th IFAC Conference on Analysis and Design of Hybrid Systems, ADHS 2021 ; Conference date: 07-07-2021 Through 09-07-2021",
url = "https://sites.uclouvain.be/adhs21/",

}

RIS

TY - GEN

T1 - A Bayesian Framework for Large-Scale Identification of Nonlinear Hybrid Systems

AU - Madary, Ahmad

AU - Momeni, Hamid Reza

AU - Abate, Alessandro

AU - Larsen, Kim Guldstrand

PY - 2021/7

Y1 - 2021/7

N2 - In this paper, a two-level Bayesian framework is proposedfor the identification of nonlinear hybrid systems fromlarge data sets by embedding it in a four-stage procedure.At the first stage, feature vector selection techniques areused to generate a reduced-size set from the given trainingdata set. The resulting data set then is used to identifythe hybrid system using a Bayesian method, where theobjective is to assign each data point to a correspondingsub-mode of the hybrid model. At the next (third) stage,this data assignment is used to train a Bayesian classifierto separate the original data set and determine thecorresponding sub-mode for all the original data points.Finally, once every data point is assigned to a sub-mode, aBayesian estimator is used to estimate a regressor for eachsub-system independently. This Bayesian approach finds acompromise between model complexity and accuracy. Theproposed method tested on three case studies. First, theidentification based on the full data set is compared withthat based on the reduced data set, over a nonlinear hybridsystem. Then, the proposed approach is compared with anexisting method from the literature, with improvedperformance over estimation error and mode assignment.Finally, the performance of the new method is tested byidentifying a complex hybrid model.

AB - In this paper, a two-level Bayesian framework is proposedfor the identification of nonlinear hybrid systems fromlarge data sets by embedding it in a four-stage procedure.At the first stage, feature vector selection techniques areused to generate a reduced-size set from the given trainingdata set. The resulting data set then is used to identifythe hybrid system using a Bayesian method, where theobjective is to assign each data point to a correspondingsub-mode of the hybrid model. At the next (third) stage,this data assignment is used to train a Bayesian classifierto separate the original data set and determine thecorresponding sub-mode for all the original data points.Finally, once every data point is assigned to a sub-mode, aBayesian estimator is used to estimate a regressor for eachsub-system independently. This Bayesian approach finds acompromise between model complexity and accuracy. Theproposed method tested on three case studies. First, theidentification based on the full data set is compared withthat based on the reduced data set, over a nonlinear hybridsystem. Then, the proposed approach is compared with anexisting method from the literature, with improvedperformance over estimation error and mode assignment.Finally, the performance of the new method is tested byidentifying a complex hybrid model.

U2 - 10.1016/j.ifacol.2021.08.508

DO - 10.1016/j.ifacol.2021.08.508

M3 - Conference article

VL - 54

SP - 259

EP - 264

JO - IFAC-PapersOnLine

JF - IFAC-PapersOnLine

SN - 2405-8963

IS - 5

T2 - 7th IFAC Conference on Analysis and Design of Hybrid Systems

Y2 - 7 July 2021 through 9 July 2021

ER -