Crystal property prediction and classification method based on attention mechanism and crystal graph volume neural network

Abstract

The invention discloses a crystal property prediction and classification method based on an attention mechanism and a crystal graph volume neural network, and the method comprises the steps: obtaining a crystallography information file and DFT calculation data of a crystal, and dividing the crystallography information file and DFT calculation data into a training set, a verification set and a test set; extracting crystal features from the crystallography information file, inputting the crystal features into a neural network, and obtaining neural network output; respectively training and verifying a constructed neural network model by adopting the training set and the verification set to obtain a prediction model and a classification model; completing the prediction of crystal properties through the prediction model, and completing the classification of the crystal properties through the classification model. The method can effectively improve the prediction and classification precision of crystal properties, consumes less time, has engineering practical value, is beneficial to realizing accurate large-scale crystal research simulation, and provides a method guarantee for development and research of new crystal materials.

Description

technical field [0001] The invention relates to crystal property prediction and classification technology, in particular to a crystal property prediction and classification method based on an attention mechanism and a crystal map volume neural network. Background technique [0002] The simulation of crystal properties is usually achieved with the help of first-principle calculations based on DFT (density functional theory), but using first-principles to screen out crystal materials with ideal properties is very time-consuming, and the calculation cost is not low . Therefore, how to realize large-scale screening of crystalline materials has become a difficult problem. With the development of computers, machine learning has gradually become an important topic in the academic field, and people are also trying to use machine learning methods to simulate large-scale crystal properties. With the continuous optimization of machine learning algorithms, the accuracy of its simulati...

AI Technical Summary

Problems solved by technology

[0007] However, the CGCNN method, as a fast and large-scale machine learning method for screening crystalline materials, has limited prediction accuracy

This is because CGCNN reduces the complexity of the network in order to improve the efficiency of the machine learning algorithm. Although the operation speed is increased, the prediction accuracy will be reduced.

And the CGCNN method defaults to 30 operating cycles (epochs), which can reduce the time-consuming process of building a model, but it also affects the fitting of the network, which will also reduce the accuracy of the prediction model.

Images