Collaborative Multiple Rank Regression for Temperature Prediction of Blast Furnace
Hongyuan Jiao, Yingwei Zhang, Chaomin Luo, Zhuming Bi
Abstract
The hearth temperature is the main factor to affect the performance of a blast furnace (BF) such as heat transfer, mass transfer and reduction, desulfurization, and pig iron composition. Therefore, the hearth temperature must be monitored and controlled to assure the quality and throughput of iron products. However, the hearth temperature can only be determined indirectly by measuring the temperature of the tuyere raceway of BF. To predict the temperature in a BF hearth promptly and reliably, a collaborative multiple rank regression (CMRR) method is proposed to predict the temperature based on the obtained images and physical variables. The proposed method is innovative in the sense that (1) images and physical variables are classified into corresponding labels to explore their correspondences; (2) design constraints are imposed on regression coefficients to avoid overfitting; (3) the locality preserving projection is adopted to tackle with diversity of big data by unifying physical variables and images in light of their similarity. The effectiveness of the proposed method is validated by the comparative studies of the CMRR with other 8 existing models using the real-world data collected from an industrial BF with a volume of 2500 cubic meters.