|
In recent years, our country's iron and steel industry has been rapid development, led to the tremendous progress in rolling technology. Beginning in the 1990s, around the set up of a number of modern mill has become the mainstream of the development of China's rolling technology. However, the degree of modernization of the rolling mill and the higher the degree of dependence on the model will be also large. Rolling model of development, maintenance and optimization of the natural will be more and more attention. For the rolling process, based on the principle of the physical and chemical reactions between the various factors in the production process, the use of traditional mathematical modeling methods, some rolling process mathematical model has been established. These models have played a huge role in rolling production. However, mathematical modeling method also has its drawbacks. This method requires the rolling process simplifications and assumptions, its accuracy can not meet the demand, to build on the experience of the engineering staff to correct their correlation coefficient. Therefore, the corresponding development of artificial intelligence-based modeling method. Artificial intelligence methods and different from the traditional method, it is not infinite explore the kind of deep-seated regular rolling process in the past, but a way to simulate the human brain to deal with those who really happening. It does not proceed from the basic principles, but based on the facts, from the large amounts of data revealed that the production pattern of the rolling process. With the growing popularity of the steel industry and in-depth, long-term production process has accumulated a rich and detailed production data, we can make use of these data, the use of artificial intelligence methods to optimize the rolling process. Therefore, it is hoped that the use of optimal control theory and technology, to optimize the improvement of the production process of hot-rolled products, improve product quality and production efficiency. The background of the iron and steel enterprises, Kunming Iron and Steel, for example, to study the hot strip rolling process. Modern metallurgical production is characterized by multi-channel processes, the rolling process from start to finish undergone major phosphorus removal, rough rolling and finishing, every step of an impact on the quality of the final product. The process of hot-rolled products rolled as a system established for this process, through this system, control of the inverse model to optimize the parameters, and then optimize the parameters into the rolling process positive model to see if its been predicted mechanical properties compared with the actual production of mechanical performance, which is more stable. BP neural network and support vector machine are two ways to establish the chemical composition and rolling parameters as input and output to the mechanical properties of the two positive control model: BP neural network hot-rolled strip is controlled models and support vector machines hot-rolled plates with a positive model; mechanical properties as input, output rolling parameters of two inverse model: BP neural network hot-rolled plates with the inverse model control and support vector machine hot strip inverse model control using inverse the positive model of the optimization parameters of the model of the production process is controlled to thereby obtain the prediction mechanical properties. Analyzed: Using our method, the effect is not ideal the BP model RM predictive value in the following performance, improvements are still needed to improve. However, on the other hand, the predicted value of BP Model RM were significantly more gathered in the vicinity of the output value of the average centerline 430MPa, ie, the parameters obtained through the inverse model variable value for controlling, the obtained mechanical properties RM more stable , volatility, and in this regard, the model has reached our objectives envisaged. Prediction error terms, 88.62% of the prediction error is less than 5%, and 99.22% of the prediction error is less than 10%, 100% of the prediction error is within 15%. The result is not satisfactory prediction error within 5%, 10% and 15% of the prediction error. Using support vector machines, in addition to the model S3T1 error of 5%, the effect is not ideal, the remaining three models are ideal. Especially model S4T2, that is, using the model of the v-SVM regression algorithm and RBF kernel function, within 5% to 100% is ideal. Visible, use the inverse model of support vector machine control, and give full play to the advantages of support vector machine for small sample, to achieve the purpose of the hot-rolled strip quality prediction. In short, whether it is using the BP neural network or using support vector machines have reached a certain degree in the hot-rolled strip quality forecasting purposes to predict the mechanical properties of hot-rolled strip products. Of course, the method still needs to continue to improve in order to improve the prediction effect.
|