Scheduling Knowledge Retrieval Based on Heterogeneous Feature Learning for Byproduct Gas System in Steel Industry. Issue 2 (2020)

• Record Type: Journal Article
• Authors: Liu, Yangyi; Lv, Zhen; Zhao, Jun; Liu, Ying; Wang, Wei
• Is Part Of: IFAC-PapersOnLine. Volume 53:Issue 2(2020)
• Page Start: 11938
• Page End: 11943
• Publication Date: 2020
• Subjects: Heterogeneous feature learning -- knowledge retrieval -- convolution neural network -- byproduct gas system; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2020.12.717
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: In the steel industry, the scheduling decisions of byproduct gas system are made based on a large number of scheduling rules. It is important to establish an effective retrieval method for scheduling knowledge, for the process of scheduling decision is complex and the amount of scheduling rules is large. In this paper, a retrieval method based on heterogeneous data feature learning is proposed, which could search the rules related to the current system state from a large number of scheduling knowledge, and help to make scheduling decisions. Considering that the data structures between the monitoring data and the scheduling knowledge texts are different, a feature learning method based on convolutional neural network is proposed to extract the key features of the scheduling knowledge, and the full-connected neural network is used to extract the corresponding working condition features from the monitoring data. Due to the features of these two kinds of data are heterogeneous, a correlation analysis model for heterogeneous features based on canonical correlation analysis is constructed, and the features are matched by the matching of maximal canonical correlation. The experimental results showed that the proposed method could effectively extract relevant data features and solve the heterogeneous gaps between the real-time data and the knowledge texts, thus effectively retrieve the corresponding scheduling knowledge in different working conditions providing supports for the scheduling work.