Unsupervised event detector for time series data in industrial Internet scenario
To solve the problem of unsupervised event detection on time series data without event labels,an unsuper-vised event extraction algorithm based on probabilistic models was proposed,which improved the efficiency and ac-curacy of the event extraction results by using pre-filtering and group voting strategies based on Generalized Likeli-hood Ratio(GLR)algorithm.Besides,a time series data clustering algorithm based on hierarchical clustering was proposed,which combined divisive hierarchical clustering with Normalized Maximum Eigengap(NME)algorithm and achieved high accuracy while estimating the number of clusters automatically.
NETL
NSTL
万方数据
