Construction of mathematical models to predict M25 and M10 coke quality indices

Introduction. Mathematical models are developed to predict M25 and M10 coke quality. The calculations are carried out specifically for each of the coke batteries of the coke-chemical production (CCP) of Magnitogorsk Iron & Steel Works (MMK). The simulation is based on the charge factors: sum of inert components OK, %; vitrinite reflectance R 0, %. These models are required to control the product quality and the optimization aimed at cost saving. The study objective is to construct adequate mathematical models for predicting М25 and М10 coke quality indices under the conditions of MMK CCP. Thus, it is assumed that MMK will obtain its own models that are highly competitive in forecast precision with the analogues used by other coke-chemical enterprises of Russia. Materials and Methods. Neural networks are used as a universal approximation for constructing mathematical models. When selecting their architecture, the authors emanated from the minimum number of neurons and the network layers. In addition, the minimization of the predictive error on a new sample was taken into account which was not used in training and testing. Research Results. The development is based on the petrographic charge factors: sum of inert components OK (according to GOST 12112); vitrinite reflectance R 0, (GOST 12113). With the help of artificial neural networks, one-dimensional mathematical models are constructed to predict impact coke strength indices of М25 and abrasion capacity of М10 (GOST 5953). The developed models are presented in a graphical form. Their predictive force is estimated. Discussion and Conclusions. In the models developed within the framework of this study, only petrographic charge factors are used. The aggregate data on technical and plastometric analysis are not taken into account. This is the basic difference of the approach presented in this paper from the models implemented for other CCP, for example, at Nizhny Tagil Iron & Steel Works (NTMK), Novokuznetsk Iron & Steel Works (NKMK), and West-Siberian Iron & Steel Works (ZSMK). Even so, the adequacy of the obtained dependences is proved. It is proposed to use them to optimize the petrographic charge factors by various optimality criteria of the coke quality indices.