In this paper, an evolved FCM-based clustering method combined with entropy theory is proposed to develop a working condition classification model for the furnace system in coal-fired power plants. To overcome the disad- vantage in beforehand determination of clustering number in basic FCM method, Silhouette index is selected as a parameter to evaluate clustering number adaptively in the process. Each time the FCM runs, the selected Silhouette
index evaluates the clustering results considering both close and separation degree. Six datasets from UCI machine learn- ing repository are used to certify the effectiveness of the evolved FCM method. Furthermore, pressure sequences from a 300-MW boiler are then discussed as the industrial case
study. Three kinds of entropy values, featured from pressure sequence in time–frequency domain, are obtained for further clustering analysis. The clustering results show the strong relationship between boiler’s load and pressure sequences in furnace system. This method can be considered a reference method for data mining in other fluctuating and time-varying sequences.

EB00000002113K

Available