A Novel Change Detecting Method for Monitoring Data Streams in Data Centers

With the rapid expansion of data centers, there are a large number of sensors in data centers to collect real-time monitoring data of various electrical equipment. It has been widely accepted that a change may indicate the fault in machine, so the changes detecting is very critical to grasp the operating status of equipment and also useful modeling and prediction of equipment operations. However, there are a great challenge to perform an online changes detecting on this kind of data stream with various patterns. In this paper, we propose a novel method to automatically implement the online changes detecting for stream data which includes three main steps: Extracting time-frequency features, Clustering them based on an improved distance metric, and Evaluating clustering results to detect change points. We applied experiments on artificial datasets and real-world datasets collected from a real large data center and proved that the proposed method can effectively solve the problem of online change detecting.