Research On The Formation And Bandgap Of Double Perovskite Oxide Based On Machine Learning

In the past few decades,with the with the continuous development of science and technology,synthesis and characterization technology have been greatly improved.Researchers have accumulated a large amount of material data.However,the traditional material research way is the trial-and-error method,that is,continuous synthesis and characterization based on experience accumulated by predecessors,until a new material meeting the requirements is found.This method does not make full use of the accumulated material data,and usually requires a lot of resources.In recent years,the rapid development of computer technology and the Internet has made the acquisition of data faster and more convenient.How to use these data to speed up the research and development of new materials has become a hot point.In this context,the application of machine learning in materials science has been developing rapidly.Based on the existing data,the models are constructed to find out the patterns contained in the data,and then the materials with target properties are designed or screened purposefully.This is an effective way to quickly discover high-performance materials.Double perovskite oxides have great potential in the field of solar cells due to their excellent stability,adjustable bandgap and low cost.However,most double perovskite oxides have a wide bandgap.The wide bandgap results in very limited light absorption and photocurrent,which greatly limits the application of double perovskite oxides in solar cells.An effective strategy to solve this problem is to find more double perovskite oxides with a bandgap in the ideal light absorption range.In addition,it is necessary to determine whether the compound is formed before studying the properties of a compound.Therefore,it is necessary and meaningful to develop a two-step machine learning screening strategy that can quickly predict the formation and bandgap of double perovskite oxides to accelerate the discovery of narrow bandgap double perovskite oxides.The main contents of the paper include the following aspects:(1)Introduce the concept and development of machine learning and its applications in the materials design and discovery,chemistry and chemical engineering.Introduce the general process of machine learning in material design and the algorithms used in this work.Introduce the structure of ABO₃ perovskite,the structure and preparation method of A₂B'B"O₆ perovskite,and the application of perovskite in solar cells.(2)Prediction models for the formation and bandgap of double perovskite oxides are established.79 double perovskite oxides with the targeted bandgaps and 75 non-perovskites are collected from published literatures,and 66 features are filled to form the original dataset.The dataset is divided into training set and test set in a 4:1 ratio.Maximum relevance minimum redundancy method is adopted with various algorithms to screen out the best feature subsets.Finally,6,4,7,11,7,and 6 features are selected to build support vector classification(SVC),random forest classification(RFC),logistic regression(LR),support vector regression(SVR),back propagation artificial neural network(BPANN),and random forest regression(RFR)models,respectively.The leave-one-out cross-validation(LOOCV)results show that the SVC model has the highest accuracy among 3 different classification algorithms,while the SVR model has the best prediction performance on the bandgap among 3 different regression algorithms.After hyperparameter optimization,the accuracy of the LOOCV and test-set validation of the SVC model reached 0.967 and 0.968,respectively.The R values of the LOOCV and independent test of the SVR model are 0.924 and 0.919,respectively.In addition,we analyzed the influence of important features on target variables.For the formation of double perovskite oxides,the chemical potential of the A-site is the most important feature,which is negatively correlated with the formation.The surface enthalpy of B'-site and the electronegativity of B"-site are the most important features affecting the bandgap,both of which are negatively correlated with the bandgap.(3)The formation and bandgap model of double perovskite oxides are used to screen narrow bandgap double perovskite oxides.6529 virtual samples are randomly generated by code programs.3590 virtual samples with tolerance factors in the range of 0.85-1.05 were initially screened.Then,3098 double perovskite oxides were screened out through the SVC model and their bandgaps were predicted by the SVR model.Finally,60 double perovskite oxides with the bandgap in the range of 1.00-1.60e V predicted by the SVR model were obtained.Among them,19 perovskites with the bandgap very close to the ideal value of 1.34 e V have the potential to be used in solar cells.In addition,data analysis shows that Fe,Ni,Sc and Co occupying B'-site and Bi,Ta,Nb,Sb,V,and Mn occupying B"-site are more likely to form narrow-bandgap oxide double perovskites.These findings can provide some guidance for the design of double perovskite oxides with narrow bandgaps.