点击次数：

论文类型：期刊论文

发表时间：2016-11-01

发表刊物：INFORMATION SCIENCES

收录刊物：SCIE、EI、Scopus

卷号：367

页面范围：311-323

ISSN号：0020-0255

关键字：Byproduct gas of steel industry; Data imputation; Non-equal-length granules correlation coefficient; Estimation of distribution algorithm

摘要：In the field of data-driven based modeling and optimization, the completeness and the accuracy of data samples are the foundations for further research tasks. Since the byproduct gas system of steel industry is rather complicated and its data-acquisition process might be frequently affected by the unexpected operational factors, the data-missing phenomenon usually occurs, which might lead to the failure of model establishment or inaccurate information discovery. In this study, a data imputation method based on the manufacturing characteristics is proposed for resolving the data-missing problem in steel industry. A novel correlation analysis, named by non-equal-length granules correlation coefficient (NGCC), is reported, and the corresponding model based on Estimation of Distribution Algorithm (EDA) is established to study the correlation of the similar procedures. To verify the performance of the proposed method, this study considers three typical features of the gas flow data with different missing ratios. The experiment results indicate that it is greatly effective for the missing data imputation of byproduct gas, and exhibits better performance on the accuracy compared to the other methods. (C) 2016 Elsevier Inc. All rights reserved.