Abstract
This paper presents a symbolization method for detecting process state change from the time series data. The proposed method comprises three processing steps as follows: (a) Peculiarity extracting of process data: the time series process data is expanded into polynomials by a peculiarity extraction filter, and peculiar points are extracted as peak points of the 2nd order polynomial coefficient. (b) Approximating by a series of vectors: the time series data is approximated by a series of vectors (line segments) which connect peculiar points in turn. (c) Comparing with dictionary: the vector series is compared with event patterns which are contained in the dictionary as the vector serieses.
The proposed method has good features as follows: (a) Calculation efficiency: the time series data is processed as a series of vectors, therefore the calculation time in the comparing step is much reduced. (b) Simplicity: the similarity measure is defined as the correlation factor between vector serieses, therefore the patterns which have similar shape but different in size are detected using an event pattern at a time.
The proposed method has been applied to an operation support system for plant and process control. The results prove that the method is practical and able to reduce the operator's work load.
