Search

The paper considers and simulates a neural network-based strip cooling process in a hot-strip mill. This process is outlined in a figure highlighting the main functional areas. A hybrid algorithm for training a radial basis neural network is described. A program was developed in the Mathcad programming unit to implement the mathematical model of the researched neural network. The input and output variables of the strip cooling process were selected, the initial data were normalized. Training and validation samples are drawn using Microsoft Access queries. A number of computational experiments were conducted for this model to obtain the optimum structure and parameters of the radial basis neural network. The results of neural network training, graphs of deviations of model values from the set ones in training and validation samples are given. The root-mean-square and relative network errors were found using the unnormalized output data. [ABSTRACT FROM AUTHOR]