|
| 铸坯质量缺陷预测的特征降维方法研究 |
| Method for feature dimensionality reduction to predict billet quality defects |
| 投稿时间：2019-12-30 |
| DOI： |
| 中文关键词: 铸坯 质量预测 最大信息系数 主成分分析 特征降维 |
| 英文关键词: billet quality prediction MIC PCA feature dimensionality reduction |
| 基金项目:国家自然科学基金资助项目(51475340). |
| 作者 | 单位 | E-mail | | 李文深 | 1. 武汉科技大学冶金装备及其控制教育部重点实验室,湖北 武汉,430081
2. 武汉科技大学机械传动与制造工程湖北省重点实验室,湖北 武汉,430081 | 1355692358@qq.com | | 容芷君 | 1. 武汉科技大学冶金装备及其控制教育部重点实验室,湖北 武汉,430081
2. 武汉科技大学机械传动与制造工程湖北省重点实验室,湖北 武汉,430081 | | | 但斌斌 | 1. 武汉科技大学冶金装备及其控制教育部重点实验室,湖北 武汉,430081
2. 武汉科技大学机械传动与制造工程湖北省重点实验室,湖北 武汉,430081 | |
|
| 摘要点击次数: 4038 |
| 全文下载次数: 2183 |
| 中文摘要: |
| 为提高铸坯质量预测的准确率,本文提出了一种基于最大信息系数(MIC)和主成分分析(PCA)的两阶段特征降维方法。采集某钢厂铸坯生产过程数据,根据冶金原理得到铸坯夹杂类质量缺陷的影响因素,构造原始特征集。第一阶段进行特征选择,使用随机森林分类器的分类准确率来评价ReliefF、IG和MIC三种算法的特征选择效果,结果显示,基于MIC度量指标选出的特征维度更低、分类准确率更高。第二阶段使用PCA方法对特征选择后的特征集进行降维,并将其与原始特征集、MIC、PCA算法的分类准确率进行比较,结果表明,本文提出的基于MIC和PCA的两阶段降维方法优于其他算法,能有效降低原始特征集的维度并提高对铸坯夹杂类质量缺陷的预测精度。 |
| 英文摘要: |
| In order to improve the accuracy of billet quality prediction, a two-stage feature dimensionality reduction approach based on maximum information coefficient (MIC) and principal component analysis (PCA) was proposed. According to the metallurgical mechanism, the influencing factors of inclusion-related billet quality defects were obtained and the original feature set was constructed. In the first stage, the feature selection was performed. The classification accuracy of random forest classifier was used to evaluate the feature selection effect of ReliefF, IG and MIC algorithms. The results show that features selected based on the MIC metrics have lower dimension and higher classification accuracy. In the second stage, the PCA method was used to reduce the dimensionality of feature set after feature selection. Compared with the classification accuracy of the original feature set, MIC and PCA algorithm, it’s found that the two-stage dimensionality reduction method based on the combination of MIC and PCA is better than other algorithms, which can effectively reduce the dimensionality of the original feature set and improve the prediction accuracy of inclusion-related billet quality defects. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
