Research Of Hot-rolling Products Quality Based On BP Neural Network Inverse Model

With the rapid development of computer artificial intelligencetechnology, it has been an important goal for the iron and steelenterprises to improve their market competitiveness and meet thecustomers’ specific demand on steel grade by this new technology.Based on the research of hot-rolling products made in an iron andsteel company, this thesis builds an inverse mapping modelreflecting the relationship between performance indicators,technological parameters and chemical composition of steel by theinverse model of BP neural networks, aiming at gettingtechnological parameters according to the given steel performanceindicators. With this purpose, this thesis includes followingcontents:1、This thesis Analysis and researches the background of thethesis and research status at home and abroad. This thesis alsoincludes introduction of the related theoires and algorithms such asthe BP neural network and adaptive inverse control theory. 2、This thesis Analysis research status and basic principle of theinternal model control theory and the design method of the linearinternal model control and the neural nonlinear internal modelcontrol. This thesis verifies the neural nonlinear internal modelcontrol has the characteristics of eliminating disturbance androbustness by experiment.3、In accordance with the BP neural network, adaptive inversecontrol and internal model control theory, This thesis has built theBP neural network inverse model with multiple input single output(MISO) based on the internal model control, and implemented theinverse mapping between the output and input variables of the BPneural networks. So the output variables can be obtained accordingto the input variables. And this thesis gives the detailed steps tosolve the inverse model.4、According to the steel samples made in an iron and steelcompany, the model is applied to the hot-rolling product qualitysystem, which allows getting rolling technological parametersaccording to given product quality performance indicators (rollingcrimp temperature), and realizes the optimization andcontrollability of rolling technological parameters. Afterexperimental verification, this thesis substitutes the crimptemperature outputted by inverse model into the hot-rolling product quality positive system to predict, within an error range of0.04,which meets the demand of the company.The next step of research work is to obtain the chemicalcomposition of the hot-rolling steel through the established the BPneural network inverse model according to a given performanceindex of steel. So the production cost will reduce, To furtherenhance the market competitiveness of iron and steel companies.