Fast Fuzzy Clustering Algorithm for Time Series in Industrial Processes

Abstract

Process industry data is often continuously measured to capture fluctuations and changes of a production processes, and this sequentially dataset can be looked as time series. We usually don’t know the correlation and classification between these data in advance. Clustering is an effective way to solve this problem, and it have been used in many fields in recent years. There are many different fuzzy clustering methods have been proposed, fuzzy c-means clustering (FCM) and fuzzy c-medoids clustering (FCMdd) are most common algorithm among them. But all of these algorithms use Euclidean distance as the similarity measure and align two times series time-by-time. These methods cannot adapt to the complex industrial environment. In this paper, dynamic time warping (DTW) is selected as the distance of the FCMdd method to adapt to complex industrial production environments, and a type of constraint is introduced to speed up it. The novel clustering method we proposed can not only speed up the clustering algorithm, but also improve the accuracy in some case. Three sets of open source sensor and simulated datasets are selected to evaluate this algorithm and achieved satisfactory results.