Industrial Process Supervisory Operation Rule Mining Based On Time Series Hierarchical Clustering Methods

In modern industrial processes,due to the extensive use of data acquisition and storage technologies such as distributed control systems(DCS),numerous supervisory operational time series data that involve rich operational information are collected.Effective time series data mining methods are expected to extract operational rules involved in the supervisory operational time series data,helping guide process operations.Tailored to the periodic or repetitive characteristics of time series data,a symbolic block processing based hierarchical clustering method is introduced to extract periodic sequences from temporal data.Applying to industrial process supervisory operations,this approach is able to help extract fragmental operational rules.This research is mainly composed of three aspects:1.In regard to periodic or repetitive time series,computational metrics for similarities among time series are investigated.Accommodated to the high dimensionality and large quantity characteristics of time series data,a method called Symbol Aggregation Approximation is adopted to reduce dimensions of time series.Additionally taking account of the continuity and directionality of industrial process supervisory operations time series data,a symbol block processing based time series data mining method is explicitly proposed.2.In order to cluster the symbolized time series,a hierarchical clustering method based on Levenshtein distance is proposed,which is able to transform the similarity matching process of time series data into the clustering process of character data,thus obtaining a variety of similar strings.Multiple similar strings form a cluster.In the case of industrial process supervisory data mining,each cluster is recognized as an operational mode.As an exemplary application,the proposed method is applied to a rectification simulation process.3.A fragmental operation rule mining method is proposed.The operation mode corresponding to process variable time series is processed by symbol tiling methods.The time series association rule mining method is used to extract the relationship between symbols along with intervals,leading to the operation rules expressed by fragmental forms.The proposed methods are consequently applied to a gasification process of coal gasification syngas scrubbing units,demonstrating the power of the contribution.