Optimized Coal Blending Strategy for High-Quality Metallurgical Coke Production Using Iranian Coking Coals: A Multi-Objective Optimization Approach with COALS Algorithm

The growing demand for high-quality metallurgical coke in Iran’s steel industry requires optimized blending strategies using domestic coals. This study uses experimental data from 18 samples across the Kerman, Tabas, and Central Alborz basins. Five optimization methods are applied: NSGA-III, QNSGA-III, RF-PSO, XGBoost-SHAP, and MOSA. A novel hybrid algorithm, COALS (Coal Optimization via Adaptive Learning System), is proposed. COALS combines QNSGA-III (exploration), PSO (exploitation), and SHAP (physical correction). Under the practical constraint of utilizing all 18 coals with a minimum 2% contribution, NSGA-III and QNSGA-III achieve CSR ≈52.3% with moderate diversity. RF-PSO achieves the highest numerical CSR (54.3%) but with very low diversity (only 2 dominant coals). A-COCO v1 achieves CSR = 55.2% (the highest overall) but concentrates 92% of the blend into five coals. A-COCO v2 offers the highest diversity (0.687) with CSR = 52.2%. COALS provides the best balance, achieving CSR = 53.8%, CRI = 21.6%, M40 = 68.5%, and M10 = 11.5%, with four dominant coals: Madanjo (16.03%), Dolati (15.83%), Saman Kavosh (14.64%), Kanjoyan (11.05%). The balance index (CSR × Diversity) confirms COALS as the optimal compromise (33.19 vs. 18.35 for RF-PSO). Industrial tests at Zarand Coke Making Complex validate these predictions with a maximum error < 1%. Under unrestricted conditions (0% minimum), COALS achieves CSR = 57.5%. Leave-One-Out Cross-Validation (R2 > 0.77) confirms model reliability. The framework provides realistic benchmarks for coke quality using domestic resources.