Improved GBRT Algorithm and Transfer Learning for Band Gap Regulation and Screening of Perovskite Materials

Abstract

As society continues to evolve, there is an increasing demand for energy. Solar energy is a clean and renewable alternative, but the photovoltaic conversion efficiency of calcite solar cells is low. Therefore, the study proposes a new material screening method for chalcocite using an improved gradient boosting regression tree (GRBT) model and migration learning. It adopts weighted averaging instead of the initial simple averaging to make complete use of the information between all the data, and at the same time introduces an adaptive step reduction to optimize the algorithm, which is fused with the support vector machine (SVM) algorithm is fused to construct a hybrid model, using hierarchical migration learning to divide the source domain data and train the model separately. Experiments showed that the astringent loss function values of the improved method were 0.115 and 0.160 lower than those of the GBRT algorithm and SVM, respectively. Moreover, and the root mean square errors and coefficients of determination for predicting the caliche band gap of the hybrid model were 0.017 and -2.2 and 0.023 and -3.7% lower than those of GBRT and SVM, respectively. The average pairwise decision error, root mean square error, and coefficient of determination of the improved transfer learning method were 0.0097, 0.0205, and -5.06% lower, respectively, than those of the ordinary method, and the running speed was 1.92 s faster than that of the ordinary method. The study screened out six halogenated bis-calcitonite new materials with band gap values in the range of [1.14-1.62] eV, and the formation energies were all below 0.05. It can be concluded that the improved method can effectively enhance the screening accuracy and speed of perovskite materials, and promote the high speed development of solar cells