Based On Artificial Neural Network Inverse Model Of Applied Research, The Production Of Hot Rolled Strip

Long-term production in the enterprise has accumulated abundant data, in which contains knowledge about hot-roll product's performance. The process of hot-roll is complex, some mechanical are indistinct, and we need operator's experience to guide the produce in practice. So, we want to improve the process of hot-roll using The Optimal Control Theory and Method.The thesis studies the production process of hot-roll product. One of the characteristic is several working procedures; the production of steel has several processes, such as Smelt, Continuous Steel Casting, hot-roll, cold rolling and so on, every process of theses has effect to the quality of the steel production. We think of the entire process of production as a whole system, and build a invert control model by the studying of the system.We use two methods to build the control model:BP neural network and SVM (Support Vector Machine). We designed two normal control models:BP neural network control model and SVM control model, whose inputs are Rolling parameters and Chemical composition and outputs are Mechanical properties; Then we built two invert control models:BP neural network invert control model and SVM invert control model, whose inputs are Mechanical properties and outputs are Rolling parameters. To join the normal control and invert control model, we get the invert model controller, and control the process using it.By the analysis, the result of the RM forecast to follow the true value of the BP model is not very well, it need to be improved. By the other hand, the most RM forecast are gather around to the average of the RM value 430MPa. It means that the control parameters which we get by the BP Invert Model used to control model, the Mechanical property RM is steadier than before. On the prediction error,88.62% are less than 5%,99.22% are less than 10%,100% are less than 15%. The prediction error less than 5% is not very well,10% and 15% are satisfactory. Using the SVM, except the model S3T1's prediction error is not very well in the 5%, the others are satisfactory. Especially the model S4T2, which use the v-SVM regression algorithm and the kernel function RBF, its prediction error is up to 100% in the 5%. So, the SVM used to control, its advantage on small data sample is satisfactory.To sum up, using the BP Neural Network or SVM, our target on the hot rolling prediction is achieved. Of cause, we should to improve our model so that we can get the better result of prediction...