Research On Self-learning Of Hot Strip Finishing Setup Modle

Improving the level of controlling the hot strip, which is always been pursued aboutthe steel rolling automatic technology. The mathematical model is always the basic ofmodern steel rolling automatic control. Improving the forecast accuracy of themathematical model in rolling process is a fundamental measure to improve the accuracyof hot strip. Although adopting the way of making better mathematical model structureand optimizing the model modulus can improve the accuracy partially,due to thevariability in the product process,it is much limited in improving model accuracy. Modelself-learning function can solve this problem better; it can correct the model in time andonline according to the situation as to follow the change in rolling process, and furtherimproving the control accuracy.On the premise of searching a big amounts of data and paper, I summarize thetendency in model self-learning home and aboard recently years, I make a deeply researchon the function of the model self-learning according to this situation, the most work in mypaper as follow:According to the problem of self-learning of finishing model in1780mm hot rolling,the paper divide the self-learning model into short-term self-learning and long-termself-learning. The point in the short-term self-learning is that how to premise the smoothmodulus in the method of exponential smoothing, proposing a multi-variable controlexponential smooth model, by the calculate of the self-learning simulation, certificatingthe rationality of the model after be premised, the result is more better than the result onlyusing the single modulus self-learning model. In the aspect of long-term self-learning, themain mission is the launch conditions of the self-learning, and how to select the strategyabout the first piece of steel after changing the layer. Firstly, joining the judge conditionsabout the degree of the changing of the standard into the launch conditions of thelong-term self-learning, ensuring the forecast accuracy, it effectively reduce the launchtimes in long-term self-learning, the continuity of the self-learning has been improved.Secondly, putting the method of tendency learning modulus, it is effectively renewedabout the equipment message belong to the self-learning modulus in layer table, improving the forecast accuracy in the layer never rolled. Finally, premising the strategyof the self-learning modulus in the layer never rolled, by reaching about the similarself-learning modulus in the rolled layer, effectively improving the accuracy of the initialself-learning modulus in the never rolled layer.