The unburned carbon content in fly ash is a critically important parameter during the boiler operation

which influences both the efficiency and safety of the boiler

therefore

it is of great significance to establish the model for unburned carbon content in fly ash. The present work employs the adaptive neuro-fuzzy inference system(ANFIS) to model the unburned carbon content in fly ash based on the real time data for a 660 MW tangentially coal-fired boiler. First of all

the initial input parameters are determined by expert knowledge and operation experience

then subtractive clustering is used to determine the initial fuzzy rules and structural parameters

the hybrid learning algorithm composed of the least square algorithm and error back propagation algorithm is employed to optimize the parameters of the ANFIS

thus the initial modelling for the unburned carbon content in fly ash is completed. Then

sensitivity analysis is used to determine the final input parameters of the ANFIS model to reduce the complexity and improve the accuracy. Finally

the model for unburned carbon content in fly ash is constructed. When applied to the test datasets

this model has high prediction accuracy

which can reflect the variation of unburned carbon content in fly ash. Moreover

compared to the least squared support vector machine(LSSVM) and typical back propagation neuro network(BP)

the proposed ANFIS model has a higher prediction accuracy and greater generalization ability in the case with enough training samples

while the LSSVM is better in the case with small training samples.