Properties Prediction Of 2D Pentagonal Materials Based On Machine Learning

2D pentagonal materials have excellent properties such as stress-dependent band gap,special topologies,and quantum Hall effects,which make them have broad application in nanoelectronic devices,photodetectors,piezoelectric materials,solar cells,spintronics,and so on.Due to the shortcomings such as long cycle and high cost of high-precision theoretical calculations and experimental measurements,machine learning model has great application prospects in the field of materials with the advantages of simple operation,short cycle and high throughput.In this thesis,we first used machine learning combined with first-principles to predict the cohensive energy and band gap of 2D penta-graphene-like materials.Using the constructed model to screen the virtual structure library,two new 2D penta-graphene-like structures with wide band gap were obtained.Then,machine learning algorithm was used to build band gap classification,regression model and structure types classification model of the 2D pentagonal materials to make a correct judgment on the relevant properties and structure types of the unknown 2D pentagonal materials.Thereby,this can reduce the time and cost required for experimental exploration and accelerate the development of 2D pentagonal materials.First two chapters,the structural characteristics,excellent performance and main application fields of 2D pentagonal materials are briefly introduced.At the same time,the research progress of current material design and property prediction methods and their respective shortcomings,as well as the research applications and advantages of machine learning in the field of materials were expounded.Finally,several machine learning algorithms involved in this thesis were introduced in detail and other commonly used algorithms were briefly summarized.In the third chapter,machine learning models were built based on simple and easily available descriptors and target values calculated by first principles,which can accurately predict the cohensive energy(E_coh)and band gap（E_g）of 2D penta-graphene like materials.Support vector machine（SVM）algorithm and genetic algorithm-multiple linear regression（GA-MLR）were used to build the regression models of cohensive energy.The R² and RMSE of test set of the SVM model was respectively 0.980 and 0.179 eV/atom,which was the best result.The band gap classification model was constructed by SVM algorithm.The classification accuracy of training set and test set was 95.6%and 83.3%,respectively.The prediction performance R² and RMSE of the test set using SVM algorithm for the band gap regression model reached 0.995 and 0.166 eV,respectively,indicating that the SVM method can obtain very accurate results for band gap classification and regression prediction.Finally,the proposed models were used to screen 720 2D penta-graphene-like structures.The first-principles calculations was used for the ten structures with the bigest predicted band gap to obtain accurate band gap,and then two new 2D penta-graphene-like structures GeN₂ and ZnS₂ with wide band gap were obtained.In the fourth chapter,SVM and random forest（RF）algorithm were used to classify the band gaps of 73 2D pentagon structure samples.The SVM model can precisely distinguish between metal and non-metal properties,and the classification accuracy of the training set and test set was 98.2%and 94.4%,respectively.Then,the regression models were performed by SVM and RF algorithm on 36 samples with E_g>0 eV.The RMSE of the test set for the GA-MLR model was only 0.247 eV.In addition,in order to help experimenters to quickly judge the structure type of 2D pentagonal materials,the selected samples were also subjected to structural classification modeling.The classification accuracy of the SVM model of the test set for the three main existing pentagonal structure types was 92.6%.Machine learning models can accurately determine the structure type of the unknown penta-structure and whether it is metallic or non-metallic in a short time,and predict the specific E_g value at the same time.This greatly reduces the time required for high-precision calculations and experimental measurements,promoting the discovery and development of new 2D pentagonal materials.