Neuro network modeling of the strip cooling process at the hot rolling station

Abstract

In the work, the process of strip cooling in the hot rolling mill of metallurgical production is considered and modeled on the basis of neural networks. This process is schematically depicted in the figure, highlighting the main functional areas. Clarified basic terms related to the above process of metallurgical production. The block diagrams of the radial-basic neural network and the Takagi – Sugeno – Kang neural network are shown, and hybrid learning algorithms are described. In the Mathcad programming unit, programs have been developed that implement the mathematical models of the considered neural networks. Input and output data for programs are respectively read and saved in Microsoft Excel files. The input and output variables of the modeled process are selected, the linear normalization of the initial data is performed. Formed training and test samples using queries in Microsoft Access. For these models, a number of computational experiments were performed, during which the optimal structure and parameters of the radial-based neural network and the Takagi – Sugeno – Kang neural network were obtained. A comparison of the trained neural networks is carried out, graphs of deviations of the training and test samples, as well as the results of their training are given. The root-mean-square and relative errors of the net-works are found from the denormalized output data. Based on the analysis performed, the main features and differences of the above neural networks are highlighted. Conclusions on the work done.