Loading…

The effects of noises on metaheuristic algorithms applied to the PV parameter extraction problem

•Self-Adaptive DE, PGJAYA and SATLBO algorithms provided the best results.•The BHCS algorithm showed the worst performance on all I-V curves.•The RMSE and Huber loss function provided the best absolute relative errors.•All parameters were accurately extracted for noise levels up to 1%.•I0 and Rsh we...

Full description

Saved in:
Bibliographic Details
Published in:Solar energy 2020-05, Vol.201, p.420-436
Main Authors: Fébba, D.M., Bortoni, E.C., Oliveira, A.F., Rubinger, R.M.
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Self-Adaptive DE, PGJAYA and SATLBO algorithms provided the best results.•The BHCS algorithm showed the worst performance on all I-V curves.•The RMSE and Huber loss function provided the best absolute relative errors.•All parameters were accurately extracted for noise levels up to 1%.•I0 and Rsh were the parameters most affected by high noise levels. This work studied the effects of noises on four metaheuristic algorithms, namely the Self-Adaptive Differential Evolution (DE), Performance Guided JAYA (PGJAYA), Self-Adaptive Teaching-Learning Based Optimization (SATLBO) and Biogeography-based Heterogeneous Cuckoo Search (BHCS) for extracting the PV parameters from I-V curves considering the single-diode model. For this task, a benchmark, simulated noise-free, and noisy I-V curves, for noise levels between 0.001 and 10% were employed, considering four objective functions: RMSE, Huber loss function, MAPE, and MAE. On the benchmark curve, the PGJAYA algorithm outperformed all the others due to a better convergence speed, while on the noise-free curve the Self-Adaptive DE surpassed all the others, obtaining the lowest absolute relative errors for all parameters. On noisy curves it was found that the objective function can significantly impact the results. In this case, the Self-Adaptive DE, PGJAYA, and SATLBO with the RMSE and Huber loss function provided the lowest errors, while the BHCS showed the worst performance, with high relative errors even for small noise levels. Also, it was found that noise affects the extracted parameters distinctly: for the Self-Adaptive DE, PGJAYA and SATLBO with the RMSE and Huber loss function, the highest relative errors for Iph were in the order of 1% for a 10% noise level; for n, in the range 7–10%; for Rs this number increased to 18.6% for a 10% noise level. Rsh and I0 showed relative errors as high as 70 and 200%, respectively, for noise levels above 5%, being the most affected parameters.
ISSN:0038-092X
1471-1257
DOI:10.1016/j.solener.2020.02.093