Application of Neural Network as Oxygen Virtual Sensor in Utility Boiler

Combustion process in utility boiler is very complicated and not fully understood until now. One operating challenge in boiler operation is the unreliable oxygen sensor could result in flame extinction of burners. An on-line virtual sensor is desirable for the unreliable oxygen sensor. This work elaborates how to build a Neural Network-based oxygen virtual sensor by make full use of the mass data available in DCS and prior knowledge for selecting model inputs. The framework can be easily transplanted to other similar applications.