古玻璃风化表面预测内部化学成分方法研究
Study on the Prediction Method of the Internal Chemical Composition of Weathering Surface of Ancient Glass
肖叙昕 1张佳怡 1谢欣欣 1鲁萍1
作者信息
- 1. 西安建筑科技大学理学院,西安 710399
- 折叠
摘要
古代玻璃在埋藏过程中的风化会导致其化学成分比例发生变化.针对古玻璃风化表面化学成分还原为内部化学成分数据方法进行研究,构建基于统计学和机器学习方法相结合的风化数据还原模型,依据表面化学成分对样本进行风化程度定义,设计无监督特征选择集成决策方法进行重要特征选取,对重要特征使用K-means模型划分风化程度类别,针对风化程度类别的统计特征及成分相关性特点建立回归、映射等风化还原模型,使用该模型还原的内部化学成分满足有效性要求,结果合理性检验使用预测误差和类别还原准确率两个指标,化学成分预测平均准确率约为67.3％,类别还原准确率约为90％.
Abstract
Weathering of ancient glass during burial can cause changes in its chemical composition ratio.The study researches the method of reducing the weathering surface chemical composition of ancient glass to internal chemical composition data,and constructs a weathering data reduction model based on the combination of statistics and machine learning methods.Then the study defines weathering degree of samples according to the surface chemical composition,and designs an unsupervised integrated decision method for feature selection to select important features.For important features,the K-means model is used to classify weathering degree categories.Regression,mapping and other weathering reduction models are established according to the statistical and component correlation characteristics of weathering degree categories.The internal chemical components reduced by using this model meet the effectiveness requirements.Two indexes of prediction error and class reduction accuracy are used in result rationality test.The average accuracy of chemical composition prediction is about 67.3％,and the accuracy of class reduction is about 90％.
关键词
无监督学习/特征选择/机器学习/K-means模型/科技考古/风化还原Key words
Unsupervised learning/Feature selection/Machine learning/K-means model/Science and technology archaeology/Weathering reduction引用本文复制引用
基金项目
国家重点研发计划重点专项(2019YFC1520200)
陕西省大学生创新创业训练计划项目(S202310703)
出版年
2024
NETL
NSTL
万方数据