Differentiated combined surrogate model and NSGA-II collaborative optimization method for burner efficiency-NOₓ multi-objective optimization

To address the inherent trade-off between combustion efficiency and NOₓ emissions in industrial burners, this study proposes a data-driven approach based on differentiated combined surrogate models collaboratively optimized with NSGA-II. Taking a 1.4 MW low-NOₓ combustion boiler as the research object, polynomial regression models for NOₓ emissions and Gaussian process regression models for efficiency were respectively established using 57,591 CFD spatial sampling data points. Subsequently, NSGA-II was employed to solve the Pareto front, and the optimal compromise solution was obtained through TOPSIS decision-making. The optimization results yielded an excess air coefficient of 1.15, a thermal load of 1.34 MW, and a flue gas recirculation rate of 0.28, corresponding to NOₓ emissions of 31.72 mg/m 3 and a combustion efficiency (here denoted combustion completeness) of 100.00%, i.e., complete combustion within the CFD resolution. Experimental validation demonstrates that the relative prediction errors for NOₓ and efficiency are 1.31% and 0.80%, respectively, confirming the effectiveness and feasibility of the proposed method in engineering applications.