Real time change point detection by incremental PCA in large scale sensor data

The article describes our work with the deployment of a 600-piece temperature sensor network, data harvesting framework, and real time analysis system in a Data Center (hereinafter DC) at the Johns Hopkins University. Sensor data streams were processed by robust incremental PCA and K-means clustering algorithms to identify outlier and changepoint events. The output of the signal processing system allows us to better understand the temperature patterns of the DataCenter´s inner space and make possible the online detection of unusual transient and changepoint events, thus preventing hardware breakdown, optimizing the temperature control efficiency, and monitoring hardware workloads.