Research On Temperature Prediction Model Of EAF With Continues Charging Based On Ensemble Learning

In recent years,EAF steelmaking technology with scrap steel as the main raw material has made great progress and has played an increasingly important role in the steel industry.Among many EAF steelmaking technologies,EAF with continues charging steelmaking technology with continuous charging,continuous preheating,continuous melting and continuous smelting has become a highly competitive technology,and more and more Adopted by the company.The endpoint temperature is an important parameter in EAF with continues charging smelting process.At present,the control of the endpoint temperature of almost EAF with continues charging relies solely on the experience of the operators.This brings about a problem that the hit rate is not high and often multiple sampling is caused,which is not only detrimental to the stability of the steel quality.The production cost is reduced,and the production rhythm of the EAF with continues charging is restricted.Therefore,the establishment of the molten steel end temperature prediction model based on soft measurement technology is particularly important,which is of great significance for improving the quality of molten steel,shortening the smelting cycle,reducing production costs and realizing steelmaking automation.In this thesis,a 120t EAF with continues charging of a steel plant in Fujian was taken as the research object,and the following model was studied for the molten steel endpoint temperature prediction model:Firstly,combined with the EAF with continues charging smelting process and its characteristics,starting from the energy budget,the endpoint temperature mechanism prediction model of the EAF with continues charging based on the energy balance of the smelting process was established.On this basis,the prediction starting point and ending point of the prediction model,the prediction time interval and the influencing factors of the endpoint temperature are analyzed.The effectiveness of the established endpoint temperature mechanism prediction model is verified by simulation experiments.Then,considering the complex characteristics of the EAF with continues charging smelting process,the mechanism model is established under certain assumptions and simplified preconditions,and many parameters in the mechanism model are obtained through experience,and only the mechanism model is used to predict the endpoint temperature of the EAF with continues charging.The temperature of the molten steel is difficult to achieve the desired effect.In this thesis,the support vector regression(SVR)is used as the data model to compensate for the modeling error of the mechanism model.The endpoint temperature hybrid prediction model of the EAF with continues charging with a parallel structure is established.The simulation experiment is effective for the established endpoint temperature mixed prediction model.Verification.Finally,to further improve the accuracy and stability of prediction,the idea of ensemble learning is introduced into the endpoint temperature prediction of EAF with continues charging.A dynamic weight combination ensemble model based on improved AdaBoost.RT algorithm is proposed for endpoint temperature prediction of EAF with continues charging.Compared with the single-use mixed prediction model experiment results,the model has great advantages in precision and generalization ability.The predicted deviation is controlled within to ±5℃ meet the actual production demand,and the hitting temperature of the endpoint temperature has been increased from 73%to 93%.Besides,for the endpoint temperature prediction model of EAF with continues charging,the model updating problem is needed in practical application.This thesis proposes a model updating method based on the Bagging multi-model idea,which is used to train a new endpoint temperature ensemble prediction mode of EAF with continues charging by training with new data sets.The endpoint temperature ensemble prediction model is then weighted and combined with the original endpoint temperature ensemble prediction mode of EAF with continues charging and used as the final model for the endpoint temperature prediction,thus realizing the update of the whole model.Compared with the traditional model updating method,the historical training results are effectively utilized,and the original data does not need to be repeatedly trained,which has the advantages of saving data storage space and reducing subsequent training time.