Industrial & Engineering Chemistry Research, Vol.58, No.11, 4518-4533, 2019
Finding Significant Qualitative Trend Combinations for Multivariate Systems from Historical Data
Data segments in similar patterns are often prerequisites for process monitoring, data mining, and decision making tasks. This paper proposes a probabilistic clustering method to automatically find significant qualitative trend combinations from historical time sequences of multiple process variables. The main idea is to obtain the histograms of decreasing, steady, and increasing qualitative trends in a data segment, and to find the qualitative trend combinations using a clustering algorithm referred to as the multifeature topic model. Numerical and industrial examples are provided to illustrate the effectiveness of the proposed method.