Research on prediction and optimization of perovskite materials based on the combination of machine learning and SCAPS-1D

Abstract In this paper, SCAPS-1D was employed to simulate the performance data of 6190 perovskite solar cells, which was used to construct a dataset consisting of 6190 groups with 10 characteristic parameters. Three ML algorithms, namely RF SVM, XGBoost, were adopted for model training, among which the RF-trained model yielded the best performance. Importance analysis revealed that the thickness of the perovskite layer was the key parameter dominating the cell performance. On this basis, 5 perovskite solar materials and structures with excellent performance were predicted. Furthermore, SCAPS-1D was utilized to optimize the thickness of the perovskite layer, carrier concentration, and work function of the back electrode. Finally, a performance breakthrough was achieved for the lead-free perovskite solar cell with the structure of FTO/TiO 2 /(FA) 2 BiCuI 6 /CFTS/Au: the PCE reached 30.02% and the FFreached 80.12%, representing an increase of 83% compared with the initial values. These results not only verify the guiding value of the energy level theory for the screening and optimization direction of perovskite solar cell materials, but also provide a reference for the design of high-efficiency and stable solar cells.