Research On Prediction Of Silicon Content In Molten Iron Based On Generative Adversarial Network

In order to achieve the comprehensive goal of high efficiency and low energy consumption in modern ironmaking process,its production mode has been gradually tending to large-scale,high efficiency and automation.The blast furnace is the most critical element in the whole ironmaking process,which main purpose is the continuous production hot metal.In order to produce high quality hot metal with lower cost,it is very important to automatically control the smooth operation of the blast furnace.The silicon content of hot metal in blast furnace can not only reflect the thermal state of the furnace,but also characterize the quality of hot metal.So,establishing a reliable prediction model of the silicon content of hot metal can not only provide reasonable feedback to the automatic control of blast furnace,but also play an important role in guiding practical production.In this paper,aiming at the key problem of the prediction of the silicon content of hot metal,the generative adversarial network whice are excellent in many fields are used for datadriven modeling research.Main research contents and achievements of this paper are as follows:Firstly,in view of the key problem of how to model the silicon content of hot metal in generative adversarial network,a prediction model based on conditional generative adversarial network is proposed.By designing appropriate objective function,network structure and distribution metric for conditional generative adversarial network,the generative adversarial network is successfully applied to the prediction of the silicon content of hot metal.In this process,the dynamic process of blast furnace is simulated by using the conversion characteristics of one distribution to another.Experimental results show that the prediction performance of the proposed method is better than that of the classical support vector machines and multilayer perceptrons.Then,it is easy to make trend prediction error when carrying out single regression modeling in actual prediction.Classification model aims at accurately predicting trend,but it cannot well reflect the detailed information of data.In order to solve this problem,this paper analyzes the prediction confidence of single regression model for the first time,and proposes a prediction model of the silicon content of hot metal based on quadrilateral generated adversation network,which combines the single regression problem with the single classification problem by using a four-role game.The experimental results show that the proposed model can not only improve the regression performance indexes,but also improve the trend prediction performance of the relevant regression performance indexes.By improving the confidence degree of the prediction of molten iron silicon content,the prediction of molten iron silicon content is more accurate.Finally,combining Docker,Rancher and front-end visual development platform,a "prediction system for molten iron silicon content" was established and the experiment was carried out on blast furnace of Liuzhou that use data-driven method to provide good guidance for blast furnace production and promote further development of industrial Internet...