Research On Cold Rolling Force Prediction Model Of Iron And Steel Based On Machine Learning

Rolling force is a crucial parameter in the cold rolling process of steel strip.Its prediction accuracy will directly affect the quality of the final product,effectively reduce the head and tail length of the strip,and improve the utilization of raw materials.In addition,the magnitude of the rolling force also determines the setting of the roll gap,which has a direct relationship with the stability of the rolling process.The rolling production site environment is complex and there are many external interference factors,which make the rolling process have variability and non-linear characteristics.The traditional mechanism model has a simple structure and many assumptions.It has a narrow application area and cannot meet the high-precision prediction requirements of rolling force.In order to improve the rolling force prediction accuracy of cold rolling mills,this paper uses the basic rolling theory,combines the mechanical model with neural networks and intelligent optimization algorithms to build the model,and uses field data to perform simulation experiments.The Bland-Ford-Hill formula model is a common rolling force mechanism model.First,combining the Bland-Ford-Hill formula,the parameters of the deformation zone are analyzed,and the variables that affect the rolling force are determined.Then,a neural network model is established,and the variables affecting the rolling force are used as model inputs to predict the rolling force.Compared with the mechanism model,the neural network model reduces the difficulty of parameter setting and avoids tedious formula calculation.For shallow models such as ELM,their network expression ability is limited by the depth of the network and cannot effectively complete the fitting of complex functions.In addition,the shallow network is suitable for small sample data and has poor processing capacity for the massive data generated during the rolling process.In order to meet the rolling force forecasting requirements under large data sets,a deep neural network model is established to forecast the rolling force.In order to solve the problem that deep networks are difficult to train,this paper adopts a combination of unsupervised andsupervised methods to train the network.At the same time,in order to extract valid information from the data,the deep sparse auto-encoder is used to complete the unsupervised training of the model.In order to filter out the noise in the collected data on site and further improve the rolling force prediction accuracy,a new deep network structure was established.The improved contrastive divergence algorithm is used to train the network,and the gradient error and direction error in the parameter update process are corrected to accelerate the convergence speed of the network.The Relu function is selected as the activation function to avoid the gradient dispersion caused by the saturated activation nonlinearity of conventional activation functions such as Sigmoid.