A Hybrid Method Combining Genetic and Nelder-Mead Algorithms for the Interpretation of Electrochemical Impedance Data - Application to Proton Exchange Membrane Fuel Cells
Electrochemical Impedance Spectroscopy (EIS) is a technique commonly used for characterizing electrochemical systems such as fuel cells, supercapacitors or batteries [1]-[3]. Modeling of the experimental results is mostly done under the form of equivalent electrical circuits (EEC) aiming to represen...
Saved in:
Published in: | 2022 International Workshop on Impedance Spectroscopy (IWIS) pp. 105 - 110 |
---|---|
Main Authors: | , , , , , , , |
Format: | Conference Proceeding |
Language: | English |
Published: |
IEEE
27-09-2022
|
Subjects: | |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Electrochemical Impedance Spectroscopy (EIS) is a technique commonly used for characterizing electrochemical systems such as fuel cells, supercapacitors or batteries [1]-[3]. Modeling of the experimental results is mostly done under the form of equivalent electrical circuits (EEC) aiming to represent the charge and mass transport phenomena and electrochemical reactions occurring at the material scale inside the individual components of the electrical components, such as resistors, capacitances, etc. It is necessary to use a numerical solver to estimate the values of the model parameters starting from experimental data. Traditional methods use deterministic algorithms, which makes the results highly dependent on the initial values chosen as starting points. Obtaining reliable and consistent results even requires sometimes to analyze separately the low, intermediate and high frequencies regions of the impedance spectra [4], [5]. In this work, we propose to use a hybrid method combining a genetic algorithm (GA) with a deterministic Nelder-Mead (NM) algorithm. The GA is used for a first estimate because of the main advantages it offers as a global optimization method: it does not require initial values for parameter estimation and is also robust and well adapted for problems with -possibly-multiple solutions. However, GA are well known to require many iterations and are therefore rather slow when some of the optimization parameters are-even partially-correlated [6]. This is the reason why the GA was combined with a NM algorithm to accelerate convergence. Both types of algorithms were taken from MATLAB standard libraries. This hybrid GA-NM method is used mainly to interpret the data measured on Proton Exchange Membrane Fuel Cells (PEMFC), which are generally modeled with Randlestype circuits or Transmission Line Models [7], depending on the shape of the spectra at high frequencies. Such EEC include at least 5 and sometimes up to 12 parameters. The results show that the combined GA-NM method converges toward satisfying solutions in most of the cases, with low least-square residuals, including when the range of parameter values is unknown and/or when the sensitivity of some of them is low. |
---|---|
DOI: | 10.1109/IWIS57888.2022.9975125 |