KIT | KIT-Bibliothek | Impressum | Datenschutz

On Interpretability and Scheduling Space Choice for System Identification with Dynamic Takagi–Sugeno Models

Lange, Konrad; Mikut, Ralf ORCID iD icon 1; Kroll, Andreas
1 Institut für Automation und angewandte Informatik (IAI), Karlsruher Institut für Technologie (KIT)

Abstract (englisch):

Dynamic Takagi–Sugeno models are often motivated by their interpretability. However, the choice of the scheduling variables with respect to interpretability is comparatively little discussed for auto-regressive models. In contrast to many neural network and kernel method approaches, Takagi–Sugeno models can expose where in the operating space the behavior changes. The scheduling variables directly influence how the nonlinear behavior can be approximated by affine local models. Here, interpretability primarily means that the partitioning of the scheduling space follows physical expectation, as opposed to other measures such as interpretability of the local models.

This paper examines how the choice of scheduling variables affects the interpretability and accuracy of auto-regressive Takagi–Sugeno models with cluster-based membership functions. Both a synthetic and a real-world example show that interpretability and accuracy are not necessarily a trade-off under a limited budget of scheduling dimensions and local models.


Verlagsausgabe §
DOI: 10.5445/IR/1000198478
Veröffentlicht am 27.10.2026
Cover der Publikation
Zugehörige Institution(en) am KIT Institut für Automation und angewandte Informatik (IAI)
Publikationstyp Proceedingsbeitrag
Publikationsjahr 2026
Sprache Englisch
Identifikator KITopen-ID: 1000198478
Erschienen in Proceedings. 36. Workshop Computational Intelligence : Berlin, 19. – 20. November 2026. Ed.: H. Schulte
Veranstaltung 36. Workshop Computational Intelligence (2026), Berlin, Deutschland, 19.11.2026 – 20.11.2026
Verlag KIT Scientific Publishing
Seiten 117-140
Relationen in KITopen
KIT – Die Universität in der Helmholtz-Gemeinschaft
KITopen Landing Page