Machine learning enhanced model for predicting coking coal blend performance with low-rank coal additions

• The current study developed a hybrid experimental - computational framework for coal blending. • Coal blending improved fixed carbon from 48.3 wt. % to 58.7 wt. % while reducing volatile matter to 28.9 wt. %. • Metallurgical blends achieved superior coke quality with CSR values of 66.0 - 69.8 and CRI values of 17.8 - 21.4. • Gieseler fluidity improved from 12 - 50 ddpm to 377 - 686 ddpm in metallurgical blends. • Random Forest regression outperformed traditional models achieving R² up to 0.97 with mean absolute errors as low as 0.05 wt. % for sulfur prediction. Achieving blast-furnace coke specifications while increasing the proportion of low-rank coals remains a key challenge for iron and steel producers. The current study addresses this challenge by integrating pilot plant-scale blending trials with multiscale characterization and machine learning to predict coke quality from coal blend properties. Ten blends from South African and imported coals were characterizedvia proximate/ultimate analyses, Gieseler rheology, ash chemistry, and standard coke tests. Transitioning from market-type blends (1 - 5) to metallurgical blends (6 - 10) lowered ash from approximately 14.6 - 15.4 wt.% to 11.8 - 12.6 wt.% and positioned the coke firmly within the metallurgical performance window (CRI ≈18–21, CSR ≈66–70, M40 ≈59–62, M10 ≈4.9–5.5). Additivity was observed for several coal parameters (e.g., ash, selected rheology measures), whereas sulfur content and volatile matter behaved non-linearly. Against linear and quadratic baselines, a validated Random Forest (RF) regressor most accurately mapped coal blend properties to coke quality, achieving R²≈0.97 for ash, 0.91 for sulfur content, and 0.93 for volatile -matter (test-set MAE ≈0.45 wt.%, 0.05 wt.%, and 1.50 wt.% respectively). RF further predicted key metallurgical coke quality (CRI/CSR) with <5% relative MAE, enabling reliable comparison of cost–quality trade-offs across coal blends. This combined approach makes it possible to accurately predict the coke quality, encourages the efficient use of lower-quality coals, leads to cost-effective, and sustainable coke production.