Generalized simulated annealing for calibration sample selection from an existing set and orthogonalization of undesigned experiments

Abstract

Abstract Generalized simulated annealing (GSA) is an optimization procedure for locating the global optimum (maximum or minimum) of multidimenisonal continuous functions. GSA has been modified for optimization of discrete functions. Selection of calibration samples from an existing set defines discrete optimization and GSA is used to select optimal sets of calibration samples for specific analysis samples. The procedure is applied to near‐infrared spectra. When compared to using the complete set of 37 calibration samples, concentration prediction errors were reduced 50%–100% by using select sets of two to seven calibration samples. Additionally, GSA was able to improve a poorly designed experiment. GSA devised augmented experimental designs such that the overall experimental design (original plus augmented) was more orthogonal than the original.