Research On Head Thickness Prediction And Optimal Setting Of Hot Rolled Strip Finish Rolling Mill

Thickness accuracy of hot rolled strip is an important indicator of product quality.Tropical thickness accuracy can be divided into strip head thickness hit ratio and strip body thickness control deviation.The head thickness hit rate depends on the accuracy of the set model.The deviation of the thickness control of the strip body depends on the deviation between the thickness of each point of the strip body and the set value controlled by the automatic thickness control system(AGC).At present,the thickness control effect of the strip body is good.To increase the thickness accuracy percentage,the control of the head needs to be improved.The thickness accuracy of the strip head is not ideal,which not only affects the thickness accuracy of the entire strip,but also affects the working condition of the AGC.Therefore,improving the accuracy of hot-rolled head control has become a realistic and serious issue.In this thesis,a 650mm hot strip continuous rolling unit in a steel plant is taken as the research object.The head thickness prediction network based on deep learning is used to predict the thickness of the strip head,which provides a reference for the parameter setting of the mill and improves the head Hit rate.Multi-objective optimization of the head thickness setting of hot-rolled steel strip is adopted to improve the head thickness setting accuracy.The specific research content is as follows:(1)In-depth study of the rolling force model and the roll gap position model,the factors affecting the thickness of the head were analyzed and analyzed,mainly including the rolling speed,rolling force,incoming thickness and rolling temperature,etc.,so as to determine the input and output parameters of the predictor.The input and output parameters are pre-processed in two ways,one is to directly perform normalization,and the other is to perform K-means clustering before normalization.(2)A head thickness prediction model of steel strip based on DNN network was established.Through the control variable method,a large number of experiments were performed on the parameters affecting the prediction accuracy of the DNN model such as the number of hidden layers,the number of nodes in the hidden layer,and the activation function.Finally,the head thickness prediction model was obtained.Optimal parameter configuration.Use the clustered data to train two DNN prediction models to further improve the model prediction accuracy.(3)From the perspectives of structural design and parameter design,a model for predicting the thickness of strip head based on CNN network is established.The CNN prediction model is optimized through experiments to obtain the optimal configuration of the CNN prediction model.Use the clustered data to train two CNN prediction models to verify the effectiveness of the clustering algorithm.Using MAE,RMSE and other regression indicators to comprehensively evaluate the prediction capabilities of DNN prediction model and CNN prediction model.(4)From the perspective of multi-objective optimization,the rolling rules are optimized to improve the accuracy of the thickness setting of the head,thereby improving the thickness accuracy of the strip head.A multi-objective optimization model of hot-rolling finishing rolling schedule was established according to the characteristics of hot-rolling finishing rolling and mathematical model of rolling process,and a multi-objective particle swarm optimization algorithm was used to optimize the hot-rolling finishing rolling schedule.The results show that the forecasting ability of the CNN network forecasting model is superior to the forecasting ability of the DNN network forecasting model.The CNN head thickness prediction model trained using clustering data,the set thickness accuracy of the strip head thickness below 4mm is within±100μm,and the prediction accuracy reaches 99.4%.The thickness setting accuracy of the strip head with a target thickness of 4mm or more is within ±150μm,and the thickness prediction accuracy reaches 99.44%.The rolling schedule using MOPSO multi-objective optimization algorithm is superior to the rolling schedule set on site.