Research On Self-learning Of Mathematical Model In Hot Tandem Rolling Mill

With the rapid development of automation technology in hot rolling, the national economic departments need very high precision dimension mill products increasingly. People have realized that the higher the degree of the modernization of the mill is, the greater dependence of the mathematical model is needed. Obviously, the mathematical model has become the core of the rolling technology. Therefore, optimizing the mathematical model continuously and improving pre-set accuracy of rolling parameters is not only the main driving force to promote the progress of rolling process computer control system, but also a fundamental measure to ensure the dimensional accuracy of the finished strip.Based on1+3high precision hot-rolling project, this thesis has researched on modeling and self-learning of process control level’s mathematical model. The process control level is characterized by the combination of accurate physical models and neural network technology. The physical model describes the basic behavior of the process, while the neural network is a compensation for the physical model. The main contents of this thesis are as follows:(1) Study hot rolling theory, the mathematical model, and their application process. On the basis of studying the related program of process control level deeply, research the modeling process of strip temperature model and rolling force model and make a detailed analysis of the model application and self-learning process in pass schedule calculations of the finishing mill.(2) Study exponential smoothing coefficient method and neural network for model self-learning algorithm, and then take short-term inheritance used in temperature model as an example to illustrate the application of exponential smoothing coefficient method. Introduce to the basic knowledge of the neural network for model self-learning algorithm briefly, then analysis of the design process and some experiences design of neural network used for model self-learning in detail.(3) Analysis of the structure of the neural network MA4and its special training algorithm used in the production line in detail. Then take short-term inheritance used in rolling force model as an example, design the structure of neural network, collect training data, and program the neural network for model self-learning algorithm. Finally calculate the long-term self-learning coefficient of rolling force model, achieve long-term self-learning function of rolling force model.Combined with the actual situation at the scene verify the validity of the neural network for model self-learning algorithm. Clearly show that the training algorithm of neural network MA4is convergent. It can be used to model self-learning, improved the preset accuracy of the model significantly.