Three multi-fidelity data hybrid-driven strategies for multi-objective combustion optimization in coal-fired boilers using POD and XGBoost

In the context of deep peak shaving and the use of low-quality coal, combustion optimization in coal-fired boilers for safety, efficiency, and low emissions remains crucial. However, due to the lack of real-time monitoring for temperature and chemical species with high resolution, as well as the significant spatial scale differences between optimization objectives, achieving multi-objective combustion optimization that balances safety, efficiency, and low pollution remains a challenge. This study develops a multi-fidelity hybrid modeling framework that couples computational fluid dynamics (CFD) simulations with field measurements to predict high-temperature corrosion and slagging severity, NOx emissions, and exhaust gas temperature. To optimize boiler operating conditions, an enhanced non-dominated sorting genetic algorithm II (NSGA-II) is employed in conjunction with three distinct optimization strategies. The three-dimensional fields of temperature and chemical species are first predicted using proper orthogonal decomposition (POD). Based on these predictions, water wall slagging and corrosion tendencies are predicted with fuzzy comprehensive evaluation (FCE) and staged evaluation. The extreme gradient boosting (XGBoost) models, integrated with k-means clustering, are trained on field data to predict NOx emissions and exhaust gas temperature. Three optimization strategies with an improved NSGA-II and constraint boundary condition are proposed and evaluated for optimization performance and time. High prediction accuracy is achieved for temperature and chemical species via the POD model, and for NOx emissions and exhaust gas temperature using the XGBoost approach. All three optimization strategies effectively reduce NOx emissions and exhaust gas temperature, with Strategy 1 requiring the shortest time, taking 28 s. However, its impact on high-temperature corrosion is limited, with a reduction of 16.1 %. Optimization Strategy 3 provides the best optimization for water wall safety. But it requires predefining a reasonable water wall safety index (WSI) limit range and longer time of 16 min. This approach introduces a novel framework for combustion optimization that simultaneously considers operational safety, thermal efficiency, and pollutant emissions.