Reserch On Flatness Control System Of Cold Rolling Leveling Mill

In recent years,the steel industry has serious excess capacity.However,the demand for high-end cold-rolled products has been increasing year by year.Therefore,eliminating backward production capacity and improving product quality are the only choices for the steel industry.As an important quality index of strip steel,strip shape is also a hot research topic in the field of rolling.More attention has been drawn to many steel enterprises and their users.In this thesis,Benxi Steel No.3 cold rolling mill continuous annealing line six-roll CVC smoothing machine as the research object,in view of the flatness of the flatness control system,many problems exist,flatness control of the flatness control system as follows.Based on the theory of machine learning,this thesis chooses a cold-rolled flatness control system with theoretical and engineering significance for the research topic,conducts in-depth theoretical research and industrial application research,and obtains new research results.Aiming at the problems that the traditional neural network model is difficult to determine the structure,the number of iterations and the long learning time,this thesis proposes a flatness pattern recognition model based on extreme learning machine,which has the advantages of fast learning speed,simple structure and extensive generalization ability.According to the essence of flatness pattern recognition and the reciprocal of shape basic pattern,this thesis adopts the fuzzy distance difference between reciprocal patterns as the input of the pattern recognition model of limit learning machine,simplifies the structure,Identification provides a new way.Aiming at the deficiency of the static influence matrix of flatness,the dynamic influence coefficient matrix method of the flatness control is proposed by analyzing the flatness regulation ability of the rolling mill.The influence coefficient of flatness changes with the change of rolling parameters in real time.Therefore,the dynamic influence matrix needs to be calculated online to reflect the effect of each adjustment method on the variation of each component of the flatness,and then to determine the new adjustment amount.In order to obtain the changing influence coefficient online,a model of dynamic influence matrix based on extreme learning machine is proposed to solve the influence matrix.The use of extreme learning machine,training speed characteristics,online replenishment of the new rules can be carried out to regulate self-learning and regulation of library replenishment,and enhance the network's learning ability and data utilization,improve the accuracy.Finally,a closed-loop on-line control scheme based on dynamic influence matrix and extreme learning machine is designed.The simulation results show that the method of flatness pattern recognition based on extreme learning machine and the dynamic influence matrix method based on extreme learning machine improve the control speed and control accuracy and meet the requirements of industrial field.