基于机器学习与SCAPS-1D的RbGeBr3钙钛矿太阳能电池性能预测与验证

Abstract This study predicts the solar cell performance of RbGeBr 3 perovskite using Machine Learning (ML) models and validated via the solar capacitance simulator (SCAPS-1D). A DFT-derived data set from the MaterialsZone repository, containing organic–inorganic halide perovskites data, was analyzed using Scikit-learn-based models and association rule mining. A total of 443 solar cell configurations were evaluated using nine key input features to predict power conversion efficiency (PCE). Among various models which includes Random Forest (RF), Decision Tree (DT), K-Nearest Neighbors (KNN), Gradient Boosting Regression (GBR), and Extreme Gradient Boost (XGB) -the XGB model performed best. Hyperparameter tuning via improved model accuracy predicting a PCE of 30.67 % for the solar cell using the XGB model shows 1.13 % of test root mean square error (RMSE) and 0.877 of test correlation coefficient ( R 2 ) for Al/FTO/SnS 2 /RbGeBr 3 /P3HT/Ni device structure. SCAPS-1D validation closely matched ML predictions, achieving PCE of 31.76 %. These findings confirm the potentiality of RbGeBr 3 for highly efficient perovskite solar cell (PSC), underscoring the synergy between ML-based screening and theoretical modeling for the design of next-generation PSCs.