Research On Neural Network Model For Prediction Of Rolling Force

The iron and steel industry is the symbol of the national economy and national defense construction,which provides strong support for the interaction between the relevant industries of national construction and enhances the level of industrialization and modernization in China.At the same time,the development of all walks of life requires the steel industry to achieve a higher level of product quality.As an indispensable part of strip rolling,temper rolling is very important to the quality control of strip shape and strip products,which can not only improve the surface quality of the strip,but also directly affect the physical,chemical and mechanical properties,and then to meet the requirements of the subsequent stages of the specification.The research on the rolling force prediction is an important task to optimize the rolling process,improve the control level and improve the working condition.Taking the parameters affecting the rolling process of the temper mill as the research object,and based on rolling theory and neural network,in this paper,the problem of the low accuracy of rolling force prediction is analyzed.And an artificial neural network model based on ReLU is proposed to predict the rolling force of the temper mill.The following research work was carried out: The principal component analysis is used to reduce the dimensions of rolling data,and the principal components which influence the rolling force are obtained which is input into the neural network model as the input layer neurons.The rolling force of the temper mill is taken as the output layer neuron of the neural network model.In this paper,the related parameters and algorithms of hidden layers of neural network are studied by the way of grid search,and python language is used for programming.Then 2360 sets neural network prediction models of rolling force are established.Based on the above research contents and results,by means of modeling analysis,combined with the analysis and treatment of a large number of data,the number of hidden layers,the number of neurons,the propagation algorithm and the regularization methods are adjusted,the neural network model with the lowest prediction error is selected.At the same time,the experimental method can be applied to the prediction of the rolling force of the temper mill,which has certain guiding significance and reference value for the smooth production,and can be extended to other parameters prediction.