End Temperature Prediction Of Molten Steel In LF Refining Process Based On Hybrid Modeling

Ladle furnace(LF)refining has become a substantial method in secondary refining.It plays as a buffer in the steelmaking and continuous casting process.LF has the advantages of good refining effect,running reliability,etc,so it was widely used in iron and steel enterprise of our country.Steel temperature was one of the important parameters need to be controlled,it was of great significance to make sure continuous casting production smoothly,reduce consumption of raw materials and energy and improve casting quality.But it was difficult to realize the online measuring of steel temperature based on technical conditions at present which meant great difficulty to accurate control of the steel ending temperature.It was an effective way to solve the above problem that established the steel ending temperature prediction model.A steel plant of 300 tons of LF refining furnace as the research object,in order to solve the problem of using a single mechanism model or a black box model,mixed modeling method was proposed to build the mathematical model that characterized the function relation between on-line measurable variables and steel ending temperature.The main research content was summarized as follows:First of all,established ladle furnace steel ending temperature prediction model through the analysis of the energy balance.The steel temperature differential expression obtained contains three unknown functions.Refining the initial time package lining temperature distribution was solved by tapping to the refining of energy balance relationship and mathematical model of heat transfer in package lining.Secondly,using BP network fitting the function that established by mechanism model.But for the real LF refining process,it was difficult to obtain the target output of neural network,so could not identify the network connection weight and threshold by conventional BP training algorithm.This paper translated the problem of getting network connection weight and thresholded into optimization problem which objective function was the root-mean-square error of predict temperature and real temperature.The optimization problem was solved by the particle swarm optimization algorithm.So,the connection weights and thresholds of the neural network could obtain.Finally,the experimental validation.Gather 500 partial production data set from one steel mill,which 400 data was used to establish hybrid model,the rest 100 data was used to test the accuracy of the hybrid model.The hit rate that steel ending temperature prediction deviation in positive and negative five degrees was 87%.The hit rate that steel ending temperature prediction deviation in positive and negative ten degrees was 91%.In order to further reflect the advantages of the hybrid model,this paper also established the end temperature of molten steel black box model based on BP neural network.The experimental results showed that the hybrid model in terms of accuracy or aggregation degree were higher than that of the black box model.