Skip to main content

Measurement and Control Engineering – Mechatronics Working Group

Prof. Dr.-Ing. Oliver Nelles

Department of Mechanical Engineering
Measurement and Control Engineering – Mechatronics
Research

6. Classification

Least-Squares Support Vector Machine

 

Properties

  • Robustly solves high-dimensional problems.
  • Least-squares support vector machines with RBF kernels → computational efficiency.
  • Two hyperparameters: standard deviation of Gaussian RBFs and regularization strength.
  • Hyperparameters can be adjusted to minimize the leave-one-out error.
  • Leave-one-out error can be computed at no additional cost in LS-SVMs: N× speedup.

  • 2-class and 1-class versions.

     

Applications

  • Design space exploration in combustion engine measurement.
  • Description of non-convex regions in combustion engine measurements
  • Modeling the stall limit in computational fluid dynamics (CFD) simulations.


Design Space Exploration
 

f69_b01 Research Nelles

 

Next Chapter: 7. Design of Excitation Signals for Identification Back to Overview