Neural network-based combustion optimization reduces NOx emissions while improving performance

A neural network based system has been applied on 11 units with tangential-, cell-, single wall- and opposed wall-burner arrangements that have ranged in capacity from 146 to 800 MW in an advisory mode. Several sites have employed the neural network-based system for closed-loop supervisory combustion control. Boiler combustion profiles change continuously due to coal quality, boiler loading, changes in slag/soot deposits, ambient conditions, and the condition of plant equipment. Through online retraining, the neural network-based system optimizes the boiler operation by accommodating equipment performance changes due to wear and maintenance activities, adjusting to fluctuations in fuel quality, and improving operating flexibility. The system dynamically adjusts combustion set points and bias settings in closed-loop supervisory control to reduce NO_x emissions and improve heat rate simultaneously. This paper presents the benefits of applying an online, real-time neural network to several commercially operating bituminous coal fired utility boilers. The system helps reduce NO_x emissions up to 60%, meeting compliance while it improves heat rate up to 2% overall (5% at low load) and reduces unburned carbon in ash (CIA) as much as 30% through combustion optimization alone. The system can avoid or postpone large capital expenditures for low NO_x burners, overfire air boiler modifications, selective catalytic reduction (SCR), and selective noncatalytic reduction (SNCR) equipment