| In recent years,with the rapid development of artificial intelligence technology,artificial neural network method is widely used in regression prediction and control,and it gradually shows its advantages in strip finishing rolling process.With the development of hot strip rolling technology,the number of characteristic variables and data involved in this process increases.In the process of prediction,the shallow neural network model may take too long to learn,or even over-fit,and gradually cannot meet the application in the strip finishing process.Lateral spread accuracy is an important indicator of the quality of rolled steel products.The improvement of width accuracy can improve the yield of products on the one hand,and on the other hand provide a better production premise for the subsequent processing.In the rolling process,the difference between the target value of hot strip finishing mill exit width and the lateral spread value of the process is usually used as the control target of rough rolling exit width.Only by accurately setting the target width of the roughing mill can the outlet width of the hot strip finishing mill be effectively controlled.Therefore,the lateral spread prediction model is the core of the width control of hot strip finishing mill,and its accuracy directly affects the width accuracy of the final product.In order to improve the accuracy and performance of the prediction model of strip rolling spread,this paper analyzes the width control principle of the hot strip finishing mill,and puts forward a prediction model of strip rolling spread based on deep learning driven by actual production data.The specific research contents are as follows:(1)Thoroughly understand and master the formation process,classification,composition and deformation mechanism of hot strip finishing mill and lateral spread,and objectively analyze the main influencing factors and change rules of spread.The technical principle of width control and the realization process of width deviation feedback control in the finishing process of hot strip rolling are analyzed in depth,providing theoretical basis and basis for the follow-up research.(2)In view of the fact that the data collected during the actual production process of hot strip finishing mill may fluctuate greatly and be missing,the Laida criterion is adopted to eliminate the abnormal data and clean the missing and duplicate data.For the discrete class characteristic variables in the sample data,the unique heat code is used to convert the numerical type to meet the requirements of continuous distribution.The Z-score standardized method is adopted to deal with the disunity of data dimension.In combination with Pearson correlation analysis,variables larger than the threshold of the given correlation coefficient are extracted.To provide reasonable and reliable sample data for the subsequent extensive prediction model.(3)Thoroughly understand and master the implementation principle of machine learning and deep learning methods and their application in regression prediction,and use MATLAB software as the platform to write the program code of data preprocessing methods such as data cleaning,unique heat coding,correlation analysis,etc.Combined with machine learning and deep learning technology,the prediction model of lateral spread based on BP neural network,support vector machine regression,short-term and short-term memory neural network and convolution and short-term and short-term memory hybrid neural network are developed respectively.With the help of parameter optimization selection and control variable method,the model combination and related performance parameters are optimized to improve the prediction accuracy and efficiency of the model.(4)By comparing and analyzing the prediction accuracy of each model,the significant advantages and characteristics of the deep learning method in Lateral spread prediction for hot strip finishing mill are summarized.The results show that the prediction accuracy and various error performance indexes of the deep learning method are better than those of the conventional machine learning and empirical formula,which provides a reliable model for the width control of the finishing process of hot strip finishing mill,and provides theoretical guidance for the improvement of the lateral spread deviation feedback system. |
