Real-time prediction and optimization of NOx emissions using artificial intelligence and online combustion data

Abstract

The study presents an innovative approach aimed at effectively reducing NO_x emissions in boiler operations. The online detection systems for coal property and carbon content in fly ash were developed to provide data for real-time optimization. Through an in-depth investigation into the relationship between boiler parameters and NOx emissions, a predictive model for NOx was established, with real-time detection data serving as key inputs. Notably, compared to the multiple linear regression model and BP (Back Propagation) neural network, support vector regression model exhibited superior prediction accuracy, with a training error of only 0.81 %. Furthermore, integrating real-time detection data into the model significantly enhanced prediction accuracy and stability, with an average prediction error of only 1.06 %. The study introduces a novel optimization strategy by considering the secondary air ratio in the main combustion zone as one of the optimization variables. Results demonstrated that incorporating this variable led to more effective reduction in NOx emissions compared to scenarios without it. Specifically, at high and medium loads, NOx concentrations decreased by 201.91 and 77.86 mg/m³, corresponding to reductions of 45.96 % and 20.17 %, respectively. Additionally, the study employed an improved genetic algorithm for combustion optimization, where the inclusion of relatively optimal solutions in the initial population yielded significant reductions in NOx emissions without compromising boiler efficiency, as confirmed through on-site validation. Overall, the developed device and methodologies provide robust support for real-time combustion optimization in boiler systems, offering promising prospects for mitigating NOx emissions in industrial settings.