Effects of Crossover Operators on Genetic Algorithms for the Extraction of Solar Cell Parameters from Noisy Data

Abstract

This study analyzed the accuracy of solar cell modeling parameters extracted from noisy data using Genetic Algorithms (GAs). Three crossover operators (XOs) were examined, namely the Uniform (UXO), Arithmetic (AXO), and Blend (BXO) operators. The data used were an experimental benchmark cell and a simulated curve where noise levels (p) from 0 to 10% were added. For each XO, the analysis was carried out by running GAs 100 times and varying p and population size (Npop). Simulation results showed that UXO and AXO suffered from premature convergence and failed to provide parameters with good precision even with very high Npop, although they provided good fitting. In all analyzed cases, BXO outperformed UXO and AXO and the results showed that it can compete with the most efficient methods. For the benchmark curve, BXO reproduced the best RMSE found in the literature (0.7730062 mA) while providing the exact values of the parameters and a very low RMSE (1E-13) for the clean curve (p=0). For noisy curves, the errors of the extracted parameters were smaller than 10% for p lower than or equal to 6%. For higher values of p, the errors were smaller than 30%.