A framework of clustering based on W-EFC with updating strategy for power plant air preheater monitoring.

This paper proposes a clustering framework with updating strategy in data stream for working condition classification and monitoring in a power plant air preheater system. A weight factor is added into the basic entropy fuzzy clustering method (EFC) for the statistical characteristics. Moreover, a framework is proposed to deal with data stream with updating block and off-line block. Four real-life data sets from UCI Machine Leaning Repository are adopted to evaluate the effectiveness (outlier detection rate and false alarm rate) of data mining framework with weighted EFC (W-EFC). These experiment results verify that the proposed framework dealing with data stream based on W-EFC has high clustering capacity and decrease noise data influence to a certain extent. Then, a field data set of the air preheater system is obtained for power plant application analysis. The clustering results can represent different working conditions and different working levels during long running time and thus can be taken as guidance for real-time operation monitoring. [ABSTRACT FROM AUTHOR]