Prediction Of Thermoelectric Properties And Lattice Constants Of Crystalline Materials Based On Machine Learnin

In recent years,machine learning has made remarkable achievements in various fields.As a general data processing technology,machine learning provides a new way for solving problems in various fields,and provides tools and methods for scientific research innovation.In this background,machine learning provides a new idea for accelerating materials research:finding regularity by using machine learning algorithms in material data and give guidance for experiments,to improve the efficiency of material research and design.Design a lattice constant prediction model and a space group classification model by using machine learning method.Using machine learning methods to explore material properties has a profound impact on improving the efficiency of material research and development,which plays a key role for promoting manufacturing transformation and industrial upgrading.The main work and innovations are as follows:Experiments were performed using data provided by the Materials Project database.Using tools such as application programming interfaces,7,700 pieces of power factor data for crystalline thermoelectric materials and 125,276 pieces of crystal structure data for lattice constant and space group prediction were obtained.For the acquired data,Magpie features are generated from the chemical formula of the material.In addition,a new feature extraction method is proposed,which can extract information about the number of atoms in a unit cell as a feature for machine learning model training.This feature can be used directly as a feature of a machine learning model,or it can be combined with features such as Magpie to train a machine learning model.Use random forest algorithm,XGBoost algorithm,deep neural network algorithm,and support vector machine algorithm to build a machine learning model for power factor prediction of crystalline thermoelectric materials.It is determined that the model built by the random forest algorithm has the best performance,with a coefficient of determination of 0.643.Propose a machine learning model MLattice ABC for lattice constant prediction.Previous lattice constant prediction models can only predict the lattice constant of a certain type of material,such as perovskite materials with the chemical formula ABO₃.MLattice ABC is a general model for lattice constant prediction that can predict lattice constants for any kind of crystalline material.The model has a very good prediction effect.For the prediction of the cubic lattice constant a,the coefficient of determination is 0.973.Propose a machine learning model for space group prediction of crystalline materials.We build the model by random forest algorithm and Magpie feature.The model has a good classification performance.For cubic crystals,the Accuracy of space group prediction is0.961.