Research On Properties Of Inorganic Phosphor Materials Based On Machine Learning

Inorganic phosphor materials have a wide range of applications in display,lighting,and solar cells due to their special electronic structure and excellent luminescence properties,and are receiving close attention from more and more materials researchers.In the research of inorganic phosphor materials,experimental research methods as well as theoretical simulation and computational methods（e.g.,first-principles methods）face challenges such as long research cycles and low efficiency,while machine learning methods have become an important means for researching material properties due to their powerful predictive capability and good generalization ability.Band gap,an important physical parameter in the energy band structure,affects the luminescence properties of inorganic phosphor materials.The formation energy is a quantitative indicator of the thermal stability of the material and determines the ease of material synthesis.Therefore,in this study,the main factors affecting the band gap and formation energy of inorganic phosphor materials are investigated by using machine learning methods to predict the band gap and formation energy.This work is valuable as a guide for the synthesis and preparation of new inorganic phosphor materials.（1）Research on prediction of band gaps of inorganic phosphor materials based on machine learning.The A_αB_βO_γ-type ternary oxide data sets is built based on the materials database Materials Project and then feature selection and data preprocessing are carried out.Six machine learning regression models:lasso regression（Lasso）,kernel ridge regression（KRR）,gaussian process regression（GPR）,random forest regression（RFR）,support vector regression（SVR）and gradient boosting regression（GBR）are built to predict the band gaps,and the validation analysis on the test set are implemented.By comparing the fitting results of different models,it is found that the band gap values predicted by most models（GPR,RFR,SVR and GBR）were in good agreement with the DFT-calculated band gap values.Among them,the GBR model has the best prediction effect and the highest prediction accuracy of 82.2%.Then,GBR is used to give the importance scores of each feature descriptor to find out the important feature descriptors affecting the band gap of inorganic phosphor materials,and it is found that the molar heat capacity and metallic valence are the important factors affecting the band gap.This conclusion is verified by the model interpretation tool Shapley Addiction Explanation（SHAP）.A qualitative explanation of the relationship between two important factors（molar heat capacity and metallic valence）and the band gap is presented.（2）Research on prediction of formation energies of inorganic phosphor materials based on machine learning.Based on the Materials Project database,the Magpie feature descriptor set is uses to characterize the formation energies of nitrides,oxides and nitrogen oxides,and then four machine learning models:GBR,SVR,RFR and K-nearest neighbor（KNN）are established to verify the test set and obtained good prediction results.The GBR model is the most effective in prediction,with an accuracy of 94.5%.GBR and SHAP are used to find out the important feature descriptors affecting the formation energy of inorganic phosphor materials.It is found that the important factors affecting the formation energy include electronegativity,space group,band gap,and valence electrons.The relationship between the important factors and the formation energy are interpreted.The formation energies of 32 UCr₄C₄-type inorganic phosphor host compounds are predicted using the trained model and compared with the calculated data in the Materials Project database,and they show a good agreement with each other.