Study On Prediction Of Desulphurization Of Molten Steel In LF Furnace And Production Scheduling Model

In the development of modern steel metallurgy,secondary refining is an important way to ensure the quality and practicability of steel and has been widely used in China's steel enterprises.The LF refining process can improve the quality of steel and expand the variety of steel.In recent years,with the continuous development of science and technology,the requirements for steel quality in various industries have become stricter.High-quality steels are required to have high strength,good low-temperature toughness,cold forming properties and good weldability.Sulfur is one of the most harmful elements affecting the performance of steel,and steel with more sulfur content is more brittle.Due to its hazard to metal materials,desulfurization has become an important part of steel metallurgy.The previous desulfurization process relies too much on the mechanism model.Due to the complicated desulfurization process,there are many influencing factors and many parameters difficult to obtain,which makes it difficult to accurately control the end sulfur content.In this paper,a steel mill LF refining furnace is taken as the research object,and the intelligent method and hybrid modeling method are proposed for the prediction of the sulfur capacity and sulfur content of the molten steel endpoint.The method of real-time learning is incorporated into the modeling process of the sulfur content prediction model.Meanwhile,the refining cycle was forecasted and added to the steelmaking-refining-continuous casting scheduling model to achieve better results.The main work of this paper is as follows:Firstly,combined with the LF refining furnace desulfurization process and its mechanism model,the improved LS-SVM method is used to predict the sulfur capacity,an important parameter in the desulfurization process.Sulfur capacity is a key parameter to measure the desulfurization capacity of synthetic slag and is an important basis for how to select the slag system.At present,the model for predicting sulfur capacity at home and abroad mainly focuses on the mechanism model.However,many parameters in the mechanism model cannot be measured,which often makes the prediction impossible,or the predicted value has a large error with the actual value.The improved LS-SVM method is applied to predict the sulfur content of the end point of the slag system in the refining process.This method overcomes the problems of small data volume and difficult measurement of parameters in the smelting field.The single-use mechanism model has some problems that some parameters cannot be accurately calculated,and the model is difficult to be adopted by the production site.However,the single-use of intelligent models has the disadvantages of over-reliance on data,lack of process guidance,and difficulty in improving forecast accuracy.In view of the above shortcomings,this paper combines the traditional mechanism modeling ideas with intelligent modeling and proposes a hybrid prediction method based on improved AdaBoost.RT method for static terminal sulfur content prediction.The weak learning machine(LS-SVM)was integrated by AdaBoost.RT method,and the weak learning machine was integrated into a strong learning machine to predict the parameters in the sulfur content prediction model,and then substituted into the mechanism model for calculation to obtain the prediction value of terminal sulfur content.This hybrid model can overcome the difficulty in measuring the parameters of pure mechanism model,the lack of process guidance of pure "black box" model,and the excessive dependence on data,so as to achieve complementary advantages.In the production process,there are often some interference factors,such as:changes in raw materials,technological changes and environmental factors,and so on.Since the sampling data at historical moment are used as the sample data of the soft test model,the working condition can only be accurately described at that time,but the description of the current working condition is not necessarily accurate.All these changes will lead to the decrease of accuracy and increase of prediction error of the previously established prediction model.In view of these dynamic characteristics,this paper proposes a hybrid terminal sulfur content prediction method based on real-time learning and online update for dynamic sulfur content prediction.The terminal sulfur content model is updated online by real-time learning method,so that it can adapt to the interference of various factors and maintain the stability of the model.This method can effectively use historical training results,do not save historical data,save storage space,reduce the time of subsequent training,improve training speed.The steelmaking-refining-continuous casting production process scheduling scheme can effectively balance the production rhythm of each production equipment on the premise of meeting the production requirements of the product.The scheme can reduce unnecessary energy consumption in the production process and improve production efficiency,thereby achieving the purpose of reducing production cost and improving the competitiveness of steel products.The LF refining process is of great significance for achieving high precision and low cost in the steelmaking process.In actual production,depending on the type of steel,the sulfur content and temperature to be achieved are different,which requires consideration of changes in the refinery cycle and composition and temperature constraints.Based on the intelligent algorithm,this paper takes the target sulfur content and the target molten steel temperature as an input,predicts the change of the refining cycle and combines with the scheduling model to optimize the scheduling model of the whole steelmaking-refining-continuous casting process.The simulation experiment proves that the method can solve the problem that the scheduling model parameters are not updated in time due to the change of the refinery cycle and meet the needs of actual production.