Font Size: a A A

Application Of Machine Learning In Performance Prediction Of Two-dimensional Nanomaterials

Posted on:2022-11-07Degree:MasterType:Thesis
Country:ChinaCandidate:W J XuFull Text:PDF
GTID:2481306779490924Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
The discovery of graphene in 2004 has aroused great attention to the research of other new two-dimensional(2D)nanomaterials.However,different 2D nanomaterials have different properties,so their application fields are also different.The zero band gap property of graphene limits the application of graphene in semiconductor nano-devices,while the indirect band gap of molybdenum disulfide is suitable for electrical and optoelectronic devices.Therefore,fast and accurate prediction of the performance of a given 2D material structure is the basis to accelerate its practical application.Here,we focus on the prediction and identification of two important properties of band gap and magnetism of 2D nanomaterials.The traditional first principle calculation needs very high computational cost in predicting material properties,and machine learning(ML)can predict and identify quickly and effectively.Although ML method has been used to predict the properties of materials,it is rarely used in 2D nanomaterials.Therefore,this paper will use ML method to study the band gap and magnetic prediction of 2D nanomaterials.Part one,we use machine learning to predict the band gap of 2D nanomaterials.In this study,based on the calculation of two-dimensional nanomaterials database(C2DB),four machine algorithms are applied to predict the band gap of two-dimensional nanomaterials,including gradient boosting decision tree(GBDT),random forest(RF),support vector regression(SVR)and multilayer perceptron(MLP).C2DB contains a large number of material attributes(features).Firstly,we select the appropriate material attributes and form a new database from a specific combination of attributes,and then divide it into training set and test set according to a certain proportion.The training set is used for modeling and the test set is used to test the effect of the model.Gradient enhanced decision tree and random forest are more effective in predicting the band gap of 2D nanomaterials,R~2>90%,and the root mean square error(RMSE)are 0.24 e V and0.27 e V,respectively.In contrast,the R~2of support vector machine and multilayer perceptron is>70%,and the RMSE is 0.41 e V and 0.43 e V,respectively.Finally,when the band gap calculated without spin orbit coupling is used as a feature,the RMSE of the four ml models is greatly reduced to 0.09 e V,0.10 e V,0.17 e V and 0.12 e V respectively,and the R~2of all models is>94%.These outcomes show that the band gap of 2D nanomaterials can be acquired fast and accurately by ML means.Part two,we use machine learning to recognize the magnetism of 2D nanomaterials.In this study,we also use the four ml algorithms in the first part to identify the magnetism of 2D nanomaterials based on C2DB.GBDT and RF are more effective in identifying the magnetism of 2D nanomaterials,F1>90%,and AUC values are 0.91 and 0.93,respectively.In contrast,F1 of support vector machine(SVM)and MLP are>80%,and AUC values are 0.85 and 0.90,respectively.These figures indicate that machine learning can quickly distinguish whether 2D nanomaterials are magnetic or not.Finally,the research of this paper is summarized and prospected.
Keywords/Search Tags:Machine Learning, Two-Dimensional Materials, Bandgap Prediction, Magnetic Recognition
PDF Full Text Request
Related items