Research On New Flatness Pattern And Intelligent Prediction Of Flatness Based On Big Data

Flatness is the key quality index of plate and strip,and the core of flatness control system is model system.The new generation of artificial intelligence theory provides a new opportunity for the research of intelligent model of flatness control.Guided by the new generation of artificial intelligence theory,based on industrial big data and aiming at practicality,this paper will study the intelligent model of flatness control.The flatness pattern recognition model and flatness prediction model are mainly studied to improve the accuracy and stability of the model and solve the shortcomings of the previous flatness control intelligent model.The specific research contents of this paper are as follows:Firstly,the characteristics of industrial cold rolling mill production data are analyzed and preprocessed,and the data set required for model training is extracted.In order to eliminate the change of input and output parameters caused by the change of strip width,a flatness channel standardization method based on piecewise linear interpolation is proposed.Secondly,the basic principle,main function,training strategy and evaluation index of self encoder are introduced.The dimension reduction model of flatness is established by using trestle self encoder and sparse self encoder respectively,and the calculation results of the two models are compared to find the best model structure.The network parameters of the bottleneck layer of the self encoder are extracted as the new basic flatness mode,which lays the foundation for the subsequent flatness intelligent prediction model.Finally,the lag between input and output parameters is analyzed and solved.Based on the deep learning theory,three neural networks with different structures(DNN,CNN and LSTM)are established for flatness intelligent prediction,and the performance of the three flatness prediction models is analyzed to obtain the best flatness intelligent prediction model.Combining the idea of dimensional compression of flatness data by sparse self encoder,the flatness prediction model based on LSTM neural network is reconstructed by using the compressed flatness data,so as to further improve the accuracy and generalization ability of the model.The research results of this paper have theoretical and practical significance for further improving the flatness control accuracy of cold rolled strip and the intelligent degree of industrial cold rolling mill.