CNN-based estimation of series resistance in photovoltaic cells from electroluminescence images with application to output power prediction

Abstract

The estimation of series resistance in photovoltaic (PV) cells is a crucial parameter that significantly influences their efficiency and overall performance. This study proposes a novel methodology to predict the slope of the current–voltage (Ⅰ–Ⅴ) curve of a PV cell in the first quadrant, where this slope (the electrical conductance) is directly associated with the series resistance of the cell. By leveraging artificial intelligence techniques, a convolutional neural network model has been developed to estimate this slope from electroluminescence (EL) images of the cells. The model was trained on a dataset consisting of EL images of PV cells with artificially induced defects, together with the corresponding slope values derived from the cells' Ⅰ–Ⅴ curves. Furthermore, this work presents a second model that combines the slope information and EL images to improve the prediction of the maximum power point (MPP) of a PV cell, surpassing previous approaches that rely solely on EL imagery. Both models demonstrated low error rates across multiple evaluation metrics, evidencing their accuracy and robustness. Additionally, comparative analysis with other machine learning methods highlights the competitive performance of the proposed approaches. These contributions provide promising tools for enhancing the assessment and diagnosis of PV cell efficiency and reliability, potentially leading to improved performance and increased longevity of photovoltaic systems.