Hot Strip Thickness Prediction Based On Deep Belief Network

As an important indicator of strip quality,the automotive,mechanical manufacturing,electrical and electronic industries have increasingly strict requirements on strip thickness accuracy.Optimizing product thickness accuracy is an important way to improve product quality.The thickness accuracy of the hot-rolled strip is mainly affected by the finishing rolling unit.During the finishing rolling process,the rolled product is deformed by the rolling force.There are many factors that affect the thickness during the deformation process,and they are coupled with each other and have serious nonlinearities.The traditional mathematical model for predicting the thickness of the steel strip omits and simplifies many influencing factors in actual production during the modeling process,so the effect of using the traditional mathematical model for error prediction is not satisfactory.In recent years,with the application of machine learning algorithms in the field of metallurgy,machine learning algorithms such as support vector machines and artificial neural networks have been widely used in the study of strip thickness prediction.However,due to the limitations of traditional shallow learning models in practical applications,these methods perform nonlinear time series regression prediction in a complex background,and high-dimensional feature extraction will not be complete,dimensionality reduction is incomplete,complex functional relationships are difficult to characterize,and multi-step prediction Major issues such as poor performance.Therefore,in order to learn the mapping relationship in high-dimensional and complex data,and then complete the regression prediction of the data,it is necessary to introduce a deep structure model to improve the prediction accuracy of the thickness of the hot strip continuous rolling.In this paper,a least squares support vector regression model for extracting data features through deep confidence networks is established,and grid search and particle swarm optimization are used to optimize the relevant hyperparameters of the model.By collecting real-time field data of Tangshan Ruifeng Steel's 950 mm hot strip rolling production line,using Matlab2017 to write a program,the BP neural network model(BPNN),the least square support vector machine model(LSSVM),the deep confidence network-least squares Support vector machine model(DBN-LSSVM)three strip thickness prediction models were trained and offline simulated.The simulation results show that the prediction model based on DBN-LSSVM has good learning ability and generalization.The average relative error of the prediction of the DBN-LSSVM model reaches 0.71%,and the prediction accuracy is higher than the traditional BP algorithm with 5.51% prediction error rate and 2.88% prediction The error rate LSSVM algorithm has been significantly improved,and the thickness prediction model has good application prospects in production practice.