Evaluation of Noisy Coherent Anti-Stokes Raman Spectra by Evolutionary Algorithms

Coherent Anti-Stokes Raman Spectroscopy (CARS) is frequently the method of choice for non-intrusive temperature measurements in combustion systems. The temperature determination requires a comparison of measured spectra with theoretically calculated ones. Conventional gradient-based least squares fitting of experimental data with a library of theoretical spectra usually leads to sensible results, as long as the spectrum shapes behave well and the noise level is low.The investigation of a 1 MW coal dust burner yielded only very noisy data that showed the limits of applicability of the gradient-based approach. Therefore in this paper a novel fitting approach based on evolutionary algorithms for such spectra is presented. The applied algorithm is explained. Temperature evaluation results, both conventionally and evolutionarily determined, are given.