[hal-01088599] A comparison of techniques to get sparse rational approximations for linear fractional representations

The objective of this paper is to stress that the size of a Linear Fractional Representation (LFR) significantly depends on the way tabulated or irrational data are approximated during the prior modeling process. It is notably shown that rational approximants can result in much smaller LFR than polynomial ones. Accor-dingly, 2 new methods are proposed to generate sparse rational models, which avoid data overfitting and lead to simple yet accurate LFR. The 1 st one builds a parsimonious modeling based on surrogate models and a new powerful global optimization method, and then translates the result into a fractional form. The 2 nd one looks for a rational approximant in a single step thanks to a symbolic regression technique, and relies on Genetic Programming to select sparse monomials. This work takes place in a more general project led by ONERA/DCSD and aimed at developing a Systems Modeling, Analysis and Control Toolbox (SMAC) for Matlab.



from HAL : Dernières publications http://ift.tt/1pxeyHF