Data Driven Based Research On The Method For Predicting The Width Quality Of Hot Rolled Strip

Control and optimization of complex industrial processes has always been an important research topic in the field of system engineering and also the key for enterprises to improve product quality and save energy and reduce emissions.Precisely establishing the classification model of strip width quality prediction,intuitively interpreting it and extracting the key process characteristics to guide the actual rolling process are of great significance for improving the competitiveness of enterprises.In this paper,the hot-rolled strip production process and data analysis methods for the hot rolled strip width quality prediction are conducted in-depth study.The main research contents and achievements are as follows:1)The existing research results in the field of hot rolled strip width control are analyzed and some practical problems faced by the current hot rolled strip width control technology are summarized.Then,the thesis introduces the common analysis methods of data mining and the classic machine learning classification algorithm.2)Aiming at the actual production data of a steel mill in the process of hot strip rolling,an mechanism based analysis and pretreatment is made.Based on the technical background of the hot rolled strip,preliminary analysis of the raw data is carried out,and the missing values and outliers in the data are cleaned,making it a more regular data set.Subsequently,the Synthetic Minority Oversampling TEchnique(SMOTE)is used to deal with the imbalance of positive and negative samples in the data.Selectively create more information about minority type of samples at the boundary between the two types of samples and avoid the possible errors caused by randomly adding samples and effectively reduce the over-fitting of the classification models.3)A prediction method for the width of hot rolled strip is proposed.Based on the statistical principle and the actual hot rolling production experience of the factory,the key features that are clearly associated with the strip width quality are found out,and based on this,9 new interactive process features are constructed.Given the large number of redundant variables in the feature set,the 0-1 coding scheme based(Binary Particle Swarm Optimization)BPSO algorithm select 36 important process features for Support Vector Machine(SVM)classifier Then,the SVM algorithm was used to study the steel strip samples which have been analyzed by the technology background and the feature engineering.The width of the unknown strip samples was predicted by the model.The test set reached 94.27%accuracy rate,proving that the establishment of the width quality prediction model in the future production has a good promotional capacity.4)A learning based SVM model rule extraction algorithm is realized.A set of process rule sets extracted from the SVM model which can be used to the width quality prediction can effectively improve the width control accuracy of the system in the future hot rolled strip manufacturing.The accuracy of the prediction is 87.98%,verifying the universality of the extracted process rule set in the hot rolled strip production process,and then assist the computer system to make the production decision quickly and efficiently.In this paper,based on the hot-rolling process and the accumulation of historical data of hot-rolled strip steel production,the methods of building a hot-rolled strip width quality prediction classification model and the rules extraction of key process sets are proposed,which is helpful to the on-the-spot supervision and control optimization of the process.It’s also beneficial to strengthen the existing decision-making function of strip width intelligent control system on the actual production site.