Publication date: 11th March 2026
In this study, the optimization of the production process for perovskite solar cells—a new generation photovoltaic technology—was carried out using machine learning algorithms with slot-die and air-knife techniques. A dataset created from experimental studies was used to train six machine learning models. Among these models, the XGBoost algorithm performed best, achieving a root mean square error (RMSE) of 0.6737 and a Pearson correlation coefficient of 0.8952. Using this model, power conversion efficiencies (PCE) for various parameter combinations of slot-die (coating gap, shuttle velocity, dispense rate) and air-knife (coating gap, shuttle velocity, air pressure) methods were predicted, and perovskite solar cells produced according to these predictions achieved a PCE of 12.739%. The architecture of the perovskite solar cell comprises a FTO layer deposited on a glass substrate, a compact TiO₂ layer for ETL, a perovskite layer (1M MAPbI₃) prepared in 2-methoxyethanol, a Spiro-OMeTAD layer for HTL, and a gold layer used as the metal contact. This result represents an improvement of 11.37% compared to the highest PCE values previously obtained in experimental studies.
