Application of artificial neural networks and regression for analysis of chemical steel reheating

Abstract

Metallurgical processes belong to complex physical-chemical processes theoretically described by means of multidimensional generally nonlinear dynamic systems with different transfer lags in their structure. Before realization of these systems control requested by practice it is necessary to execute their structural and parametric identification. As these processes are very complex, all exact relations for their mathematical description are not known so far. Some metallurgical systems are practically non-described so far (black box), further described only partially (grey box), while only a little of them are described almost fully (white box). Determination of internal structure of insufficiently described systems is done by means physical modelling, by measurement if important data and subsequently by means of regression analysis or artificial neural networks applied to measured data. There is certain chance to determine a proper system internal structure at system identification by means of statistical analysis (i.e. to come from black box to grey box or from grey box to white box), though this approach is knowledge and time-consuming. Identification by means of artificial neural networks enables rather external system description (i.e. black box models creation), when we get an acceptable accordance between real and modelled outputs, i.e. so called output estimation (prediction). This approach is thus more suitable for control than for identification itself. Contribution deals with a possibility of prediction of a temperature after a steel chemical heating on device of integrated system of secondary metallurgy by means of regression analysis and artificial neural networks and with a comparison of both of these approaches.