BEGIN:VCALENDAR
VERSION:2.0
METHOD:PUBLISH
BEGIN:VEVENT
ORGANIZER;CN=ESTAD 2023:mailto:info@metec-estad.com
LOCATION:Room 01
SUMMARY:Data mining approach to coke quality prediction: linear and gradient boosted models from production data
DESCRIPTION:	Being able to accurately predict metallurgical coke properties for different parent coal blends is an important part of optimizing an integrated steelmaking plant production costs and ensuring blast furnace process safety. However, it is challenging to balance the economic pressures to use different coals and different blend compositions, the need for minimal product quality and the complexities of the coke-making process. The present work aims to construct coke quality models based on traditional analyses of constituent coals and some process parameters to enable better prediction for a wide range of blend characteristics. To this end, historical production data for process parameters and qualities of coke, coal and petcoke documented in a by-product coking plant were collected and processed using data mining and machine learning techniques. Attempts were made to model coke strength after reaction (CSR), stability and hardness indices by combining domain knowledge with feature engineering, feature selection algorithms and proposing three different models for each quality index - multiple linear regression, gradient boosted decision trees with XGBoost and gradient boosted linear trees with Light GBM. Cold strength models, especially those created using gradient boosting techniques, attained reasonable predictive ability. It is concluded that despite the inaccuracies surrounding industrial data collection, data mining and machine learning techniques provide a viable and promising framework for coke quality modeling in mature and well-documented processes.
CLASS:PUBLIC
DTSTART:20230615T100000
DTEND:20230615T102000
END:VEVENT
END:VCALENDAR