Double Perovskite Screening And Structure-activity Relationship Study Based On High-throughput Computing And Machine Learning

As a star material in the field of photovoltaic,perovskite solar cell has become a hot spot for current research which has excellent electrical and optical properties.Due to the toxicity of lead and instability of materials,scientists have proposed double perovskite structure to replace the traditional lead-based perovskite.Although theoretically double perovskite exhibits better properties than lead-based perovskite in carrier life,mobility and other aspects,the efficiency of current synthetic double perovskite solar cells is still not comparable to that of lead-based perovskite solar cells.This thesis mainly explores and promotes the properties of materials in double perovskite family.The specific research contents are as follows:（1）The large number of family members have brought difficulties to traditional material screening methods.The emergence of machine learning provides new ideas and methods to solve this problem.This thesis combines machine learning and first principles calculations to develop a goal-driven method to search functional materials,aiming to select which has suitable photoelectric properties and thermal stability from tens of thousands of double perovskite materials.Among the models established by the six machine learning algorithms,the Gradient Boosting Decision Tree performs best,with a determination coefficient of 99.8%and a mean square error of 0.057.According to the predicted results,11 excellent double perovskite materials were successfully selected out of 16400 candidate materials through multi-condition combination screening and band gap and stability verification based on first-principles.They are Cs₂Ag VBr₆、Rb₂Ag VBr₆、K₂Ag VBr₆、Na₂Ag VBr₆、Cs₂Ag Cr Br₆、K₂Ag Cr Br₆、Rb₂Ag Cr Br₆、Na₂Ag Cr Br₆、K₂Ag Bi Br₆、Rb₂Ag Bi Br₆、Cs₂Ag Bi Br₆,which will be used to guide the subsequent experimental synthesis.Combining the high efficiency of machine learning and the accuracy of first principles calculations compensates for their respective shortcomings and improves the efficiency of finding new materials.In addition,this method also has a certain general applicability and can be extended to the screening of other materials.（2）Machine learning greatly improves the efficiency of material search.If the calculation of photoelectric conversion efficiency can be further realized,the pros and cons of candidate materials can be evaluated more intuitively.And the SCAPS software can achieve this goal.Cs₂Ag Bi Br₆ and other classical double perovskites have promising photoelectric conversion efficiency in theory,but it has not exceeded 3%in experiments.In this thesis,SCAPS software is used to simulate the J-V curve of Cs₂Ag Bi Br₆ solar cell to explore the reasons for the low photoelectric conversion efficiency of Cs₂Ag Bi Br₆ solar cell.The results show that the defects and excessive band gap are two main reasons.In view of the defects,the simulation value of photoelectric conversion efficiency can be improved from 2.94%to 3.88%by adjusting the layer thickness.After alloying,the double perovskite material has a suitable band gap,and the simulated value of photoelectric conversion efficiency is up to 8.92%.