Real-time forecasting of key coking coal quality parameters using neural networks and artificial intelligence

High quality coke is a key raw material for the metallurgical industry. The characteristics of the coal have a significant influence on the parameters of the coke produced and, consequently, on the valuation of coal deposits and the economic assessment of mining projects. Predicting the quality of coking coal allows for the optimisation of production processes, including the planning and management of operations and the early detection of quality problems. In this study, using the principles of a smart mine, it is proposed to determine the quality of coal based on the combination of mining and geological conditions of mineral deposits and its quality indicators. Possible interrelationships between the quality of the coal in the deposit and the characteristics of the final product have been identified. A neural network is used to determine the priority of individual indicators that have a significant impact on the quality of coking coal. An important part of the research is its practical implementation in the conditions of the Jastrzebska Spilka Weglowa SA. Qualitative and quantitative parameters of coking coals were obtained for each mine of the region by the method of sampling and statistical processing of data such as: degree of metamorphism, thickness, deviation of volatile substances, presence of phosphorus, ash content, etc. For their evaluation, the Group Method of Data Handling was used to compare the factors of quality indicators depending on the priority of influence on the final characteristics of the coking coal. Based on the results obtained, it is shown that not all coal quality indicators have a significant impact on the quality of the final product. The study shows that it is possible to predict the main indicators (CRI - Coke Reactivity Index, CSR - Coke Strength after Reaction) of coke quality using neural networks based on a larger number of coal quality parameters and to eliminate parameters that have virtually no influence on the value of the final product. This method can also be used to improve the results of economic valuation of a deposit and to better plan exploration and mining operations.