Research On The Feature Selection Method Of Input Parameters For Converter Steelmaking Carbon Temperature Prediction Model

The iron and steel industry is a basic industry for national economy,and converter steelmaking plays an extremely important role in it.The main task of converter steelmaking is to predict the carbon content and temperature at the end of smelting.The data of converter steelmaking process is an important factor affecting the outputs of carbon and temperature prediction model.The way to select the key features from input parameters is significant to improve the accuracy and efficiency of prediction models.Therefore,it is need to select reasonable parameters,which are from the high-dimensional data characteristics of the converter steelmaking process,as input nodes for neural network forecast model.And it is necessary to conduct in-depth analysis and research on feature selection methods based on converter steelmaking process data.This dissertation takes the actual production process data of converter steelmaking as the research object.Based on optimization and selection of input parameters of carbon and temperature prediction model for converter steelmaking,three kinds of feature selection algorithms are proposed to conduct in-depth research.First,an important contribution degree(ICD)algorithm is proposed to select the data input features of the steelmaking process,the validity of the ICD algorithm is verified by the results of the carbon and temperature prediction based on the back propagation neural network model.Then we assess the deficiencies of the ICD algorithm,and put forward a method of combining grey relational analysis(GRA)and ICD(GRA-ICD algorithm)to select the input characteristics of the steelmaking process.The applicability of the GRA-ICD algorithm is validated by the carbon and temperature forecast results of the steelmaking end point,and it is proved that the GRA-ICD algorithm has certain advantages over the ICD algorithm.Finally,a vector similarity method is used to weight improve the GRA-ICD algorithm(ie,the improved GRA-ICD algorithm is proposed).This method is applied to the selection experiment and analysis of the input characteristic parameters of converter steelmaking forecasting.The validity of the weighted improvement of the GRA-ICD algorithm is verified by the carbon and temperature forecasting accuracy of the steelmaking endpoint.The experimental results show that the improved GRA-ICD algorithm has certain advantages over ICD and GRA-ICD algorithms,and the forecast accuracy is higher than other feature selection methods adopted in this paper.This further validates the effectiveness and applicability of improved GRA-ICD for the selection of input characteristic parameters for carbon and temperature-prediction models for steelmaking destinations.In summary,the purpose of the feature selection methods studied is to obtain the key parameters which affecting the outputs of carbon and temperature.Especially the improved GRA-ICD algorithm,can provide a methodological reference for converter steelmaking prediction models.It provides a certain theoretical and experimental basis for the application of intelligent prediction and control of converter steelmaking point in actual industrial production.