Segmentation of historical process data based on fuzzy clustering algorithm

Abstract

The segmentation of historical process and business data is an important data-mining task, as the resulted segments are used for building predictive models, effective storing and querying of historical databases, and rule-searching. In this paper a tool is presented which can analyse the data collected during the operation of technologies and can determine the time periods in which the behaviors of the analysed variables are similar. This tool is based on fuzzy clustering. The effectiveness of the proposed algorithm is presented in the case studies based on real data sets from the polyethylene factory of the TVK Ltd. The results prove well that this tool can be applied to distinguish the typical operational periods i.e. to qualify the operation of the technology posteriorly based on the measured data.