Premium
Genetic Algorithm‐Based Improvement to Equivalent Circuit Analysis of Experimental Data in Electrochemical Impedance Spectroscopy
Author(s) -
Kowski Marek,
Palys Marcin J.
Publication year - 2021
Publication title -
chemelectrochem
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.182
H-Index - 59
ISSN - 2196-0216
DOI - 10.1002/celc.202100778
Subject(s) - algorithm , genetic algorithm , equivalent circuit , dielectric spectroscopy , electrical impedance , noise (video) , limit (mathematics) , least squares function approximation , non linear least squares , population , sequence (biology) , nonlinear system , computer science , mathematics , mathematical optimization , electrochemistry , voltage , statistics , engineering , estimation theory , physics , electrical engineering , chemistry , electrode , artificial intelligence , mathematical analysis , estimator , image (mathematics) , sociology , biochemistry , quantum mechanics , demography
The paper discusses implementation of a hybrid algorithm for the optimization of equivalent circuit (EQC) parameters in the processing of electrochemical impedance data. Similar to other works, we use the genetic algorithm (GA) to find approximate values of parameters, which are further refined by a complex nonlinear least squares (CNLS) method, e. g. Levenberg‐Marquardt. This removes the important drawback of CNLS: the need to supply optimized starting values that are close to optimal ones. We propose the sequence of GA operations and values of GA parameters. In comparison to other works, our algorithm is designed to simplify connection of its GA part to existing CNLS procedures in electrochemical software packages. The algorithm was tested with datasets generated for both simple and complex EQCs, with and without added noise. No starting values were needed and the allowed ranges for EQC parameter variations were broader than found in usual experimental practice. For GA population as small as 300 individuals and the limit of only 500 GA cycles, more than 85 % of runs were successful, while in many cases the success rate was 100 %, also for smaller populations.