Use of artificial neural network for prediction of characteristics of metallurgical coke.

The artificial neural network is used for simulation of the coke quality indicators: CRI – coke reactivity index and CSR – coke strength after reaction with carbon dioxide. It is noted that the task of determining the quality characteristics of coke, CRI and CSR, can be attributed to insufficiently formalized task in which there are many influencing factors that cannot be taken into account explicitly. The solution of such tasks is possible using the artificial neural networks. Bilayer artificial neural network was used for CRI and CSR simulation. Logistic function was used as a compressive function. Backpropagation and the steepest descent methods were used for neural network training. Information on batch mixture materials for coke and by-product process of JSC EVRAZ NTMK for 10 months was used as input information. Initial data included proportions of mixing components and characteristics of each component. The effect of the number of neurons on the simulation results was studied. The possibility of predicting CRI and CSR for five (three months) during the neural network training for the previous five (seven) months was studied. The difference between the experimental and calculated data is in the range of 6–18 %. [ABSTRACT FROM AUTHOR]