Research On Band Gap And Stability Of Inorganic Non-Lead Halide Double Perovskite Solar Cell Materials Based On Machine Learning

In the past decade,perovskite solar cells have experienced rapid development.However,stability and lead toxicity remain the major obstacles to their large-scale application.Halide double perovskites have gained increasing attention as an effective solution.Compared to traditional ABX₃perovskites,halide double perovskites have a general formula of A₂B¹⁺B³⁺X₆,involving a greater combination of chemical elements.It is evident that machine learning,with its flexibility and lower cost compared to traditional trial-and-error methods or DFT simulations,is gradually becoming a necessary pathway for discovering new materials.At the same time,the rapid development of perovskite solar cells has accumulated a large amount of experimental and theoretical data for machine learning.Therefore,employing machine learning is not only crucial but also meaningful in finding halide double perovskite materials with suitable bandgaps and stability,as well as uncovering key features that influence material properties.The main content of this paper can be summarized in the following aspects:（1）A predictive model for the bandgap and stability of halide double perovskite materials was established.A total of 632 halide double perovskite structures were downloaded from the Materials Project material database to obtain their bandgap and Ehull values,which were used to evaluate the light absorption ability and stability of the materials.Additionally,45 atomic features were collected to form the initial dataset.Seven regression algorithms were compared:Support Vector Regression（SVR）,Random Forest Regression（RFR）,XGBoost Regression,Artificial Neural Network（ANN）,Gaussian Process Regression（GPR）,Bayesian Ridge Regression（BRR）,and Gradient Boosting Regression（GBR）.Their predictive abilities for the bandgap and stability parameters were assessed.The feature importance analysis was performed to customize feature subsets for different algorithms using recursive feature elimination.A 10-fold cross-validation was conducted,and the results showed that the ANN model had accurate predictions for both the bandgap and Ehull values.After hyperparameter optimization,the model’s generalization ability was evaluated on the test set.The bandgap prediction model achieved an R²value of 0.988,with mean absolute error（MAE）and root mean squared error（RMSE）values of 0.131 e V and 0.172 e V,respectively.The stability prediction model achieved an R²value of0.959,with MAE and RMSE values of 0.017 e V/atom and 0.022 e V/atom,respectively.（2）The analysis of important features on the target variables reveals the following influences.For the bandgap of halide double perovskites,the electronegativity of the B¹⁺site element is the most important feature and negatively correlates with the bandgap.The ionic radius of the X site element and the octahedral factor（O_f）also have a significant impact on the bandgap.The former is negatively correlated,while the latter is positively correlated with the bandgap.Regarding the stability of halide double perovskites,the ionic radius of the A site element has the greatest influence and negatively correlates with the Ehull value.The tolerance factor（T_f）,the ionic radius of the B³⁺site element,and the electronegativity of the B¹⁺site element are the next important features.Specifically,the electronegativity of the B¹⁺site element is positively correlated with the Ehull value.（3）The band gap and stability prediction model of halide double perovskite was used to select the solar cell materials with the best band gap and stability.The element replacement method was used to generate 11,880 candidate materials and fill in key features.The trained model is used to predict the band gap and stability parameters.First according to 0.8<T_f<1.2,0.4<O_f<0.7,2729 samples were initially screened.391 combinations were selected according to the band gap variation within 1.00～1.60e V.The Ehull threshold was set to 0～0.05 e V/atom,and 28 structures with suitable band gap and good phase stability were obtained.In addition,the analysis shows that the selection of B¹⁺element Na,Ag,TI,B³⁺element As,In,Bi,Sb may obtain a relatively stable halide double perovskite.These findings provide ideas and guidance for the development of halide double perovskite materials with photovoltaic application prospects.