Research On The Prediction Model Of Rolling Force For Tandem Cold Rolling Process Based On Data-Driven

The accuracy of the rolling force during the cold rolling process determines the shape,dimensional accuracy,and quality of other steel strip products.Existing theoretical models of rolling force introduce too many assumptions,which,together with the difficulty of measuring deformation resistance in the models,make it difficult to adapt the existing mechanistic models to the complex working conditions of a deeply non-linear,strongly coupled and uncertain cold rolling process.In recent years,advanced technologies represented by industrial big data and artificial intelligence have developed rapidly.To increase the rolling force prediction’s precision,this paper uses a data-driven approach to establish a deformation resistance prediction model and further develops a Light GBM based rolling force prediction model.The main research elements of this paper are as follows:First of all,the analysis of the theoretical rolling force model provides the relevant factors affecting rolling force and deformation resistance.Based on the actual production data,the method takes the deformation resistance as an optimisation variable,and by establishing the objective function,the theoretical value of rolling force calculated by substituting the measured data matches the actual value of rolling force,and the particle swarm optimisation algorithm is used to solve the objective function to obtain the actual value of deformation resistance.On the basis of this,high quality data is provided for data-driven models by means of missing value processing,outlier detection based on box plots and isolated forests and Min-Max normalisation.Secondly,to address the problem that the theoretical value of deformation resistance affects the accuracy of rolling force,a support vector regression(SVR)based deformation resistance prediction model is constructed.Twenty-one features are selected as model input variables,taking into account the hot rolling process temperature,alloy composition and other factors.To address the problem that the accuracy of the support vector regression model is affected by penalty factors and kernel function parameters,Grey Wolf Optimization(GWO)algorithm is used to optimize parameters and K-fold cross validation is used to train the model.Compared with other models,the GWO-SVR model has stronger prediction performance;furthermore,the prediction results of the models are compared for different steel grades,and the average relative error of the GWO-SVR model within ±5% accounts for 98.4% of the data.Finally,a Light GBM based rolling force prediction model is proposed to address the problem of numerous factors influencing rolling force and the inability of accuracy to meet the complex production process of cold rolling.On the basis of obtaining deformation resistance,feature selection is carried out by mutual information,which reduces the complexity of the model.The influence law of Light GBM hyperparameters on the model is analysed,and the optimal combination of parameters for the model is determined by manually adjusting the parameters to establish the rolling force prediction model.Through model comparison,the superiority of the Light GBM model in terms of prediction accuracy and generalisation capability is verified.Further,in order to improve the model operation efficiency and prediction accuracy,the parameters of the Light GBM model were optimised using Bayesian optimisation algorithm and four swarm intelligence optimisation algorithms.The results show that the optimised Light GBM model has improvement in prediction performance and model stability,and the salp swarm algorithm can control 99.5% of the data in the test set within ±5%,which is a reference for the actual production process.