Machine learning algorithms for predicting electrical performance of perovskite solar cells

Perovskite solar cells (PSCs) have undergone significant evolution in a short time, achieving an efficiency increase of approximately 22% in less than 15 years. These advancements have been driven by empirical findings from experimental cells, leading to increased costs and labor in the development of this technology. Therefore, based on these empirical findings with experimental cells, predictions close to reality can be obtained using machine learning regression algorithms. These models seek complex patterns in a series of variables that aim to describe the cell. This paper aims to compare machine learning algorithms to predict the electrical performance of a PSC using variables such as open-circuit voltage (Voc), current density (Jsc), fill factor (FF), and power conversion efficiency (PCE), based on variables classified as molecular structure, cell layers, and manufacturing processes. The methodology comprises variable selection, feature engineering, model selection, and model evaluation. For the electrical variables, a mean absolute percentage error (MAPE) of 9.3% and a mean absolute error (MAE) of 0.05 V are predicted for Voc; a MAPE of 10.4% and a MAE of 1.5 mA/cm² for Jsc; a MAPE of 10% and a MAE of 0.05 for FF. It is concluded that it is feasible to implement machine learning models to predict the electrical performance of PSCs. This integration represents an advancement in renewable energy technology by providing a predictive model that reduces reliance on costly and labor-intensive experimental procedures. The results contribute to the fields of renewable energy and machine learning by offering a tool that can lead to improved efficiency and cost-effectiveness in PSC development.