Combustion diagnosis is very important for combustion optimization of boilers. In this study

the average value

standard deviation

uniformity

coefficient of variation

and peak-to-peak value of the flame detecting signals that characterizes the combustion state of the furnace are adopted as characteristic variables. The fuzzy C-means clustering algorithm is used for combustion diagnosis. Properties of the selected features are analyzed and the acquisition cycle of the sample during the diagnostic process is researched in detail. The results show that

the variation of the clustering center in numerical value can accurately represent the practical combustion state by using the selected features. The acquisition cycle plays an important role on accuracy of the combustion diagnosis. When the acquisition cycle is 5 seconds

there is a difference in the variation trend between the clustering center in numerical value and the practical combustion state

and the diagnostic accuracy rate is only 80%. When the acquisition cycle is 10 seconds or 30 seconds

100% accuracy of the diagnostic is obtained. It may result in misdiagnosis when the acquisition cycles of samples for calculating the clustering centers is different from the acquisition cycles of samples to be diagnosed. In practical applications

it is necessary to keep the acquisition cycle consistent.