Optimized Artificial Neural Network with Improved Sparrow Search Algorithm for Predicting Bandgap and Stability of A2B 1+B 3+X6 Perovskites

Abstract

With the rapid development of solar power generation technology, chalcogenide battery materials are becoming more and more important. However, there are many combinations of chemical formulae for chalcogenide materials, and the traditional calculation methods are too costly in terms of labor and time. Therefore, an artificial neural network-based prediction model for the band gap and stability of halogenated double chalcogenide (A2B 1+B 3+X6) is proposed in the study. The structural features of the material are extracted using the Voronoi diagram method. The sparrow search algorithm is improved by using the chaotic number generator, stochastic difference variant, dynamic allocation strategy, and nonlinear inertia factor. The number of algorithmic populations is set to 50. The maximum and minimum values of the ratio of discoverers and joiners are 3:7 and 1:9 respectively. The learning factor is 1 and the warning value is 0.5. The improved algorithm is used to optimize the artificial neural network. Experiments show that the optimal fitness value of the improved algorithm is 145, 128, and 53 lower than that of GBR, SVR, and XGBoost, and the running time is 0.035 s, 0.127 s, 0.022 s, and 1.212 s lower than that of GBR, SVR, and XGBoost respectively, indicating that the improved algorithm performs better in optimization problems. The root mean square error is 0.053, which is lower than SSA, GBR, SVR, and XGBoost algorithms by 0.036, 0.019, 0.101 and 0.038 respectively. The mean absolute error and root mean square error of the model are 0.0217 and 0.0354 lower than that of the XGBoost model, respectively, and the coefficient of determination is 5.46% higher. The mean absolute error and root mean square error distributions of the model are lower than that of the XGBoost model by 0.0217 and 0.0354, and the coefficient of determination is higher by 5.46%. The six A2B 1+B 3+X6 mines selected for the study meet the requirement of band gap between [1.3, 1.4], the band gap prediction error between [-0.047, 0.009], and the stability indexes of five of the materials meet the requirement of less than 0.05. It can be concluded that the study can effectively improve the screening speed of chalcogenide materials, reduce the screening cost, and provide more promising new materials for solar power generation