Improvement of the Blast Furnace Viscosity Prediction Model Based on Discrete Points Data
- PDF / 582,858 Bytes
- 10 Pages / 593.972 x 792 pts Page_size
- 101 Downloads / 202 Views
I.
INTRODUCTION
THE slagging regime is one of the four blast furnace (BF) operation systems. As a significant metallurgical property indicator of BF slag, viscosity produces a great influence on the BF production. The main prediction methods of metallurgical melts are briefly overviewed as follows: (1) the geometric model estimate method based the regular solution theory[1–4] calculates the viscosity of multicomponent system based on each binary subsystem viscosity data; (2) the physical viscosity prediction model according to the quantum mechanics and statistical mechanical principle;[5,6] (3) the data mining and regression algorithm based on the viscosity data, such as the neural network method, the Chou model, and so on;[7–10] and (4) in view of the melts structure analysis, the model parameters are determined by using the regression method of viscosity data.[11–15] From the above research, it has been found that method 1 is unsuitable for the viscosity prediction of BF slag because of the high melting point of each binary subsystem like the Al2O3-SiO2 binary. Method 2 contains an in-depth study on the viscosity mechanism, which gives well-defined parameters but resulted in a greater estimating error on occasion because of its incomplete analysis of melts structure and insufficient revelation of the viscosity mechanism. Methods 3 and 4 are based on a large number of experimental data; thus, both of them HONGWEI GUO, Vice Professor, is with the Shagang School of Iron and Steel, Soochow University, Suzhou, Jiangsu 215021, P.R. China. Contact e-mail: [email protected] MENGYI ZHU, Master, is with the Department of Ferrous Metallurgy-Metallurgy of Iron and Steel, RWTH Aachen University, Aachen 52056, Germany. XINYU LI, Master, JIAN GUO, Doctor, and JIANLIANG ZHANG, Professor, are with the School of Metallurgical and Ecological Engineering, University of Science and Technology Beijing, Beijing 100083, P.R. China. SHEN DU, Master, is with the State Key Laboratory of Advanced Metallurgy, University of Science and Technology Beijing, Beijing 100083, P.R. China. Manuscript submitted August 10, 2013. Article published online October 2, 2014. 378—VOLUME 46B, FEBRUARY 2015
require mass viscosity data with high accuracy. Method 3, the neural network prediction method, can fit nonlinear variables; however, its fitting performance relies on the amount of nodes in the hidden layer. On the one hand, when the amount of nodes is lower, the skills of gaining information from the sample are worse, which is not enough to summarize the sample law. On the other hand, the effect of the nonregularity content, such as noise, will be magnified when the amount of nodes is larger. As a result, its process fails to be controllable and interpretative; the Chou model has shown good interpretability and is a controllable process, but there is room for improving the distance selection and extensionality prediction and so on. It seems that method 4 should have better prediction accuracy from the view of the theoretical analysis, but sometimes it introduces
Data Loading...
