E3S Web Conf.
Volume 267, 20217th International Conference on Energy Science and Chemical Engineering (ICESCE 2021)
|Number of page(s)||7|
|Section||Environmental Chemistry Research and Chemical Preparation Process|
|Published online||04 June 2021|
Prediction Of Material Properties By Neural Network Fusing The Atomic Local Environment And Global Description: Applied To Organic Molecules And Crystals
1 Beijing Advanced Innovation Center for Materials Genome Engineering, Beijing Information Science and Technology University, Beijing, China
2 Synfuels China Technology Co., Ltd., Beijing, China
Machine learning has brought great convenience to material property prediction. However, most existing models can only predict properties of molecules or crystals with specific size, and usually only local atomic environment or molecular global descriptor representation be used as the characteristics of the model, resulting in poor model versatility and cannot be applied to multiple systems. We propose a method that combines the description of the local atomic environment and the overall structure of the molecule, a fusion model consisting of a graph convolutional neural network and a fully connected neural network is used to predict the properties of molecules or crystals, and successfully applied to QM9 organic molecules and semiconductor crystal materials. Our method is not limited to a specific size of a molecule or a crystal structure. According to the calculation principle of the properties of the material molecules, the influences of the local atomic environment and the overall structure of the molecules on the properties are respectively considered, an appropriate weighting ratio is selected to predict the properties. As a result, the prediction performance has been greatly improved. In fact, the proposed method is not limited to organic molecules and crystals and is also applicable to other structures, such as clusters.
© The Authors, published by EDP Sciences, 2021
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.