Constraint-Aware Coal Blending Optimization for Coal-Fired Power Plants Using Random Forest and Coati Optimization

To address the problems of inaccurate blended coal quality prediction and difficulty in constrained coal blending optimization in coal-fired power plants, a data-driven coal blending optimization method based on Random Forest (RF) and the Coati Optimization Algorithm (COA) is proposed. First, an RF-based prediction model is developed to estimate key blended coal quality indices, including heating value, volatile matter, ash content, and sulfur content, under small-sample conditions. Second, based on actual operating data from an ultra-supercritical unit and relevant national standards, plant-specific boundary constraints for as-fired coal quality are established. Third, the RF prediction model is embedded into the COA-based optimization framework to search for blending schemes with minimum procurement cost under multiple coal quality constraints. The results show that the RF model achieves higher prediction accuracy than the conventional weighted average method for all four quality indices. In the case study, the cost-oriented optimized scheme reduces the procurement cost by 5.05% while satisfying the prescribed coal quality constraints. The proposed method provides a feasible decision-support approach for coal blending management in coal-fired power plants.