Multi-dimensional characteristic construction methods of computational materials under big data environment

Abstract

Characteristic construction is an important part of material data analysis. In the big data environment, the development of material data science will have implications for multidimensional data analysis methods. Among these, the machine learning method for multidimensional data models can be widely applied to material data types, including element ratios, atomic compositions, electronic arrangements, molecular structures, and energy distributions. For high-throughput computing materials, it is recommended in material data science to judge and extract the main characteristics influencing material properties and predict novel functional materials using the discovered laws. Consequently, we considere the characteristic construction, learning prediction, feature extraction, and high-order analysis of computational materials as the main research purposes, and construct a composite analysis model of the material system by combining data preprocessing, data mining, data evaluation, and knowledge representation as the main steps of data analysis. This demonstrates that a method to comprehensively judge the properties of materials by constructing the characteristics of materials in different dimensions is essential.

Keywords

• Material characterization
• Feature extraction
• High-order analysis
• Data mining
• Machine learning