Study on deconvolution of alpha-particle spectra using metaheuristic optimization and multivariate analysis

Abstract

To find a global optimal solution and obtain further insight into the parameters of the peak shape fitting function, metaheuristic optimization algorithms, and multivariate analysis techniques are employed to study the deconvolution of alpha-particle spectra. An improved peak shape model composed of the Bortels-Collaers function and Lévy distribution function, which aims to handle the high-energy tailing, is proposed. A newly developed metaheuristic optimization method, Bonobo Optimizer (BO) is adopted to seek optimal parameters in the peak shape function. Multivariate analysis (MVA) techniques are used to find hidden information in the shape model. Pearson's correlation tells the mutual variation relationship among parameters, while Multidimensional scaling (MDS) shows similarities of parameters through a 2D plot. Effects of parameters upon the regression accuracy are obtained via the Student t-test. Self-organizing Mapping (SOM) is utilized to mine intrinsic relations among these parameters through visual images. AM243-1 test alpha spectra example is selected to examine the proposed methodology. The improved model is more accurate when handling the high-energy tailing features. Compared with traditional gradient-based optimization algorithms, BO can find global solutions without tedious work in initial solution setting and constraint handling. More information is mined through MVA and further understanding of the peak shape function is obtained.