Abnormal Signal Detection based on Time Series Clustering

Abstract

Abnormal signals are extremely important to almost every area including industry, medical and transportation. However, the various huge quantity signals make signal detection applying to every condition a challenging problem. Anomaly detection of time series mainly has three difficulties: time sequence, high dimensionality, no common rules. This paper proposes a novel algorithm for signal detection based on temporal clustering. The algorithm uses convolution layer and Bi-GRU to reduce the dimensionality of time series data and get latent representation. Clustering with latent representation to detect types of abnormal signal can solve the problem effectively. To prove the effectiveness of the algorithm, simulation of the real signal and visualization of the result are also done properly.