Machine Learning-Assisted Study Of The Photonic Energy Of The Quasi-Two-Dimensional Perovskites

Perovskite light-emitting diodes have attracted more and more researchers’ attention due to their simple preparation method,low cost,and high color purity.The highest external quantum efficiency of red and green devices has reached more than 20%,gradually approaching the level of commercial use.However,blue perovskite materials suffer from low efficiency and poor spectral stability,limiting the application of perovskites in full-color displays and lighting.As a branch of artificial intelligence,machine learning can quickly process massive data sets and map the relationship between the variables and the result.Hence,we combined machine learning with experiments to enable the high throughput exploration of blue quasi-two-dimensional perovskite materials.The machine learning models were established and used to predict the photon energy of quasi-two-dimensional perovskite materials.Furthermore,the experiments were carried out to fabricate blue materials and verify the predicted results.The main content of this thesis is as follows.(1)Based on different machine learning algorithms,the regression model on the photon energy of quasi-two-dimensional perovskites was established.We collected relevant data on blue quasi-two-dimensional perovskite materials from literature.We extracted 7 features,including the photon energy,the proportion of Cs in A-site cations(Cs/(Cs+FA+MA),Cs ratio),the ratio of Cl ions in halogen elements(Cl/(Cl+Br),Cl ratio),and organic spacer cations.We utilized four algorithms to learn the dataset and evaluated their performance with the root mean square error(RMSE)and Pearson coefficient(r).After data mining and hyperparameter optimization,the random forest model performed best on the test set(RMSE=0.44,r =0.90).The photon energies of perovskite materials with various combinations of the variable scales were predicted using the random forest model.In addition,the effects of the solubility correlated parameter(XLog P3)of the organic spacer cations in polar solvents,Cs ratio,and Cl ratio of the perovskites on the photon energy and the related mechanisms were discussed.(2)Based on the predictions,a series of quasi-two-dimensional perovskites with different compositions were designed and fabricated,which led to the blue emissions and the validation of the predicted results.By adjusting the Xlog P3 value、Cs ratio,and Cl ratio,we prepared the quasi-two-dimensional perovskite films under different conditions.The photoluminescence spectroscopy and absorption were used to characterize the photon energies and bandgaps of the films under different conditions,which verified the reliability of the prediction results by machine learning.Meanwhile,several materials with emission peaks located at the blue wavelengths of 440-480 nm were obtained.These results prove the high potential of machine learning in predicting the photon energy of perovskite materials.