SIMULATED-ANNEALING-BASED OPTIMIZATION ALGORITHMS - FUNDAMENTALS AND WAVELENGTH SELECTION APPLICATIONS

Simulated annealing (SA) is a stochastic search method that has been applied to combinatorial search problems in chemometrics. Unlike strict iterative improvement methods, SA tolerates temporary moves to detrimental parameter configurations during an optimization. The method used by SA to decide whether or not to accept detrimental steps is a special case of a more general acceptance rule. The present work investigates the performance of various SA-type algorithms that differ only in the acceptance rule for detrimental steps when optimizing continuous or discrete problems. A method for step width modulation is introduced to overcome the poor ability of SA type algorithms to locate the exact extreme of a function. The studied search strategies are modified for the discrete problem of wavelength selection. In order to evaluate SA-type algorithms and their abilities to deal with the wavelength selection problem, two global measures of selectivity are used as criteria to determine the most suitable wavelength subset that maximizes selectivity for pure component ultraviolet-visible spectra.