Analysis And Modeling Research On Hot Metal Silicon Content Of Blast Furnace Smelting Process

Blast furnace as an upstream process of the steel production process, its direct and related carbon dioxide emissions account for 90% of the total emissions in steel industry. Energy consumption in blast furnace smelting accounts for 70% of the total energy consumption in steel industry. Therefore, the blast furnace smelting is the main potential region of energy saving and emission reduction in steel industry. The hot metal silicon content not only stands for the characterization of hot metal quality but also represents the temperature in blast furnace. For the blast furnace smelting process with high temperature and high pressure, the transferring behavior of silicon in the blast furnace is subjected to complex coupling effect from many factors. Thus, the accurate establishment of hot metal silicon content model for the blast furnace smelting process is the premise of realizing efficient control of the blast furnace, and is also a difficult problem of automation in iron making.For the key issue of the hot metal silicon content modeling, a systematic study of the data driven modeling scheme is carried out in the background of No.2 blast furnace in Liugang. Through the analysis of the complex migration behavior of silicon in blast furnace and the statistical analysis of the field operation data, key factors affecting the final hot metal silicon content and the corresponding lag time are determined. In the construction of the modeling algorithm, we analyze the time varying characteristics of the blast furnace system for the first time. Thus, the model proposed in this paper can predict the hot metal silicon content when the furnace condition fluctuates violently. Finally, we develop “ a hot metal silicon content prediction system of blast furnace” for No.2 blast furnace in Liugang. The experimental results of the system are good. The main work of this academic paper includes the following aspects:Firstly, through analyzing the migration behavior of the silicon in the blast furnace deeply, the mechanism function relationship between the various variables in the blast furnace and the hot metal silicon content is achieved in this paper. Then, its influencing factors are obtained from the analysis of mechanisation. Furthermore, through analyzing the correlation between the corresponding variables and the hot metal silicon content, the correlation size of them is obtained. Moreover, the lag time of the corresponding variable acting on the hot metal silicon content is determined. These provide a good method for selecting appropriate input variables for the data driven modeling. Therefore, it is the basis for accurately establishing the hot metal silicon content model.Secondly, aiming at the problem that the algorithm’s training time increase due to the increase of the number of hidden nodes when applying the extreme learning machine modeling algorithm to solve the complex practical problems, an improved extreme learning machine algorithm based on matrix decomposition is proposed. This new algorithm can not only retain the performance of the original algorithm, but also greatly reduces the training time. Moreover, the proposed method can be extended to many other improved extreme learning machines algorithms. It has been proved that the method has good result on modeling of the hot metal silicon content in blast furnace.Then, we firstly consider the time varying characteristics of the blast furnace system, which is caused by the variation of the internal environment during the blast furnace operation process. A variable forgetting factor stochastic gradient Wiener modeling method is proposed for the hot metal silicon content. The recursive identification method of the variable forgetting factor stochastic gradient is employed to identify the nonlinear dynamic structure of the Wiener model. This method can track the variation of the blast furnace system. Therefore, it can provide a good prediction result.Furthermore, we consider the time varying characteristics of the blast furnace system deeply. Consider the interactive change of some furnace conditions in the blast furnace smelting process, the variable forgetting factor stochastic gradient can be divided into some submodels. To solve above problem, a new algorithm named gated extreme learning machine which can not only grasp the multi-model characteristics, but also make full use of the historical data is designed. The data simulation has verified the effectiveness of the algorithm. Especially, for the furnace condition with large fluctuations in the hot metal silicon content, the algorithm can provide good prediction.Finally, based on the proposed algorithm, the blast furnace hot metal silicon metal prediction system is developed by using the LabVIEW and MATLAB hybrid programming method. The experiment carried on the No.2 blast furnace in Liugang proves the effectiveness of this system. The system can provide a good operating instruction for the blast furnace operators.