基于机器学习的PVDF基复合介质储能特性数据分析与预测  

Data Analysis and Prediction of Energy Storage Performance in Polymer Composite Dielectrics Based on Machine Learning

作　　者：冯宇 唐文昕 张天栋 迟庆国[1] 陈庆国[1] FENG Yu;TANG Wenxin;ZHANG Tiandong;CHI Qingguo;CHEN Qingguo(Key Laboratory of Engineering Dielectrics and Its Application,Ministry of Education,Harbin University of Science and Technology,Harbin 150080,China)

机构地区：[1]哈尔滨理工大学工程电介质及其应用教育部重点实验室,哈尔滨150080

出　　处：《高电压技术》2022年第5期1997-2004,共8页High Voltage Engineering

基　　金：国家自然科学基金(51807041;51977050);黑龙江省自然科学基金(ZD2020E009);中国博士后科学基金(2020T130156);清华大学电力系统国家重点实验室开放课题(SKLD20M13)。

摘　　要：近年来,机器学习作为一种新型数据分析方式,在电气、材料、化学等领域都取得了优异的成果。对储能介质材料而言,以聚偏氟乙烯(polyvinylidenefluoride,PVDF)材料作为基体,向其中加入纳米填料能够极大地增加复合介质最大储能密度。该研究利用机器学习探索并建立复合介质所含填料(微观信息)-复合介质储能性能(宏观性能)的对应关系。首先,收集165组复合介质储能特性参数建立数据库,以填充相材料的特征作为输入描述符(包括固有描述符和选择描述符);其次,对原始数据进行处理,根据复合介质的最大储能密度提升倍数划分数据集标签。为达到兼顾预测精度和准确率的目的,分别设置二分类、三分类和四分类数据集,使用3种机器学习算法对数据集进行训练;最后,将11组全新的数据输入训练模型进行验证,其中7组数据可以正确预测分类,证明机器学习方法应用在高储能密度复合介质研究中的可靠性。该研究将交叉学科的前沿成果运用在复合介质的研究领域,所建数据库与训练模型将加速高性能复合介质的发现。In recent years,machine learning,as a new way of data analysis,has made excellent achievements in the fields of electricity,materials and chemistry.In the field of the energy storage dielectric materials,the maximum energy storage density of the composite can be greatly increased by adding nanofillers to the polyvinylidene fluoride(PVDF).In this study,machine learning was used to explore and establish the corresponding relationship between the fillers(micro information)and the energy storage performance(macro performance)of the composite dielectrics.First,165 energy storage characteristic parameters of composites were collected to establish a database,and the characteristics of the filling phase material were taken as input descriptors(including inherent descriptors and selective descriptors).Then,the original data were processed and the dataset labels were divided according to the maximum energy storage density promotion multiple of the composites.In order to achieve the purpose of both prediction accuracy and accuracy,data sets of dichotomous classification,triple classification and quadruple classification were set respectively,and three machine learning algorithms were used to train the data sets.Finally,11 sets of new data input training models are verified,among which 7 sets of data can be correctly predicted and classified,proving the reliability of machine learning method in the study of high energy storage density composite media.In this study,the frontier results of cross-disciplines are applied to the research field of composite dielectrics,and the database and training model established will accelerate the discovery of high-performance composites.

关 键 词：复合介质 最大储能密度 纳米填料 机器学习 数据集标签 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TM53[自动化与计算机技术—控制科学与工程] TQ317[电气工程—电器]