A fast and precise genetic algorithm for a non-linear fitting problem

• Published in: Computer physics communications Vol. 124; no. 2; pp. 204 - 211
• Main Author: Brunetti, Antonio
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.02.2000; Elsevier Science
• Subjects: Algorithms for functional approximation; Biological and medical sciences; Computational techniques; Exact sciences and technology; Fit procedure; Investigative techniques, diagnostic techniques (general aspects); Mathematical methods in physics; Medical sciences; Miscellaneous. Technology; Numerical approximation and analysis; Physics; Radiodiagnosis. Nmr imagery. Nmr spectrometry; Radiographic measurements; Spectral methods; X-ray; X-ray radiography; Fit procedure; Radiographic measurements; X-ray; Non linear fitting; Radiography; Fitting; Genetic algorithm; Theoretical study; X ray; Measurement method
• ISSN: 0010-4655; 1879-2944
• DOI: 10.1016/S0010-4655(99)00454-3

Fitting procedures are currently used in a large set of computational problems and several algorithms have been developed. However, a complication appears when the fitting function is non-linear and non-lineariable. In this case, a Marquardt–Levenberg procedure is generally used, but it often requires interactions with the user. Here a new method is proposed which is based on a genetic algorithm technique. This kind of algorithm allows fitting in a completely automatic mode, without any manipulation over the fitting function. The algorithm developed is generally faster and more precise than traditional genetic algorithms reported in the literature. Its performances are comparable to those in the Marquardt–Levenberg algorithm technique. It has been developed as fitting method for measurements of X-ray tube response. Fitting this response is very important to avoid any patient injuries. The results obtained are reported here and compared to other genetic algorithm implementations, as well as a Marquardt–Levenberg procedure.