Reliable Behavioural Modelling
| Led by: | Prof. Dr.-Ing. Erich Barke |
| E-Mail: | markus.olbrich@ims.uni-hannover.de |
| Team: | Dipl.-Ing. Anna Krause |
| Year: | 2014 |
| Is Finished: | yes |
We develop new algorithms to generate data-based behavioural models. The models have to be compatible with the AGIAS simulator, which was developed at the institute.
The resulting models should have the following properties:
- The parameters are represented by affine forms, a special interval representation in affine arithmetic.
- The models enclose the original data.
- No prior knowledge is necessary to generate models.
We base the new algorithms on well-known support vector machines (SVMs), εSVMs and νSVMs. The parameters for the new SVMs are represented by affine forms. For the new algorithms, two new parameters have to be estimated. This is done by including additional constraints in the optimisation problem.
The resulting algorithms have been used successfully to model circuits and simplify complex component models.
