<![CDATA[Enhancing Circulating Fluidized Bed (CFB) boiler efficiency is a critical objective in industrial energy management, often hampered by the intricate, non-linear dynamics of operational parameters. This study evaluates and benchmarks three machine learning architectures—Random Forest (RF), XGBoost, and Artificial Neural Network (ANN) to develop a robust predictive model for boiler thermal efficiency using historical industrial telemetry. The analysis utilizes six key operational variables, including Air-Fuel Ratio (AFR) and Bed Temperature, for model training and cross-validation. Empirical results demonstrate that XGBoost serves as the most effective predictive framework, achieving a Coefficient of Determination ( ) of 0.495 and a Root Mean Square Error (RMSE) of 0.040, thereby outperforming RF and ANN in capturing industrial data noise through its sequential optimization and regularization mechanisms. A primary finding identifies AFR as the most influential factor, exhibiting a strong positive correlation (0.84) and consistent top-tier feature importance rankings across all paradigms. This research provides a validated data-driven methodology for real-time boiler optimization, emphasizing stoichiometric synchronization as the paramount strategy for improving thermal performance and minimizing fuel-related operational expenditures.]]>
Copyrights © 2026
