可扩展性
外推法
阴极
电压
计算机科学
过程(计算)
化学空间
膨胀的
主动学习(机器学习)
高压
机器学习
材料科学
人工智能
电气工程
工程类
数据库
化学
统计
数学
复合材料
操作系统
药物发现
生物化学
抗压强度
作者
Heekyu Kim,Jaejung Park,Minseon Kim,Jaejun Lee,Inhyo Lee,Kyoungmin Min,Seung‐Chul Lee
标识
DOI:10.1021/acs.jpcc.3c05812
摘要
Reliable databases for cathode materials are scarce and have limited scalability; this has hindered the development of a systematic methodology for exploring high-performance materials within the vast chemical space of cathode materials. Using 153,424 data points related to cathode materials, this study implemented an active learning platform that expedites the search for high-voltage materials. By exploring up to 20% of the total data, the proposed platform could discover 79.19% of samples with an average voltage within the top 1%, whereas when searching up to 50% of the total data, it could identify 99.35% of such samples. The proposed platform could effectively search for materials with high average voltages even when the available training data contained materials with a low average voltage. The predictive performance of the machine learning model trained through the active learning process exhibited a mean absolute error of 0.263 V and an R-squared value of 0.921 for the dataset not sampled during the process, demonstrating its capability for the rapid screening of high-voltage materials. Our platform can also handle extrapolation tasks, demonstrating its potential as a tool for discovering new materials outside of the training data. The proposed platform can be used for swiftly and efficiently finding materials with desirable properties, even in the presence of diverse and expansive chemical spaces for materials that are outside the available training data.
科研通智能强力驱动
Strongly Powered by AbleSci AI