Study of Anomaly Detection on Tracking of Residual Vector based on Subspace Method

Abstract

Fast and accurate anomaly detection is becoming essential in maintenance services. This paper proposes a novel detection approach based on the subspace method. To achieve high performance of the method, a part of learning data is respectively selected based on the distance between observed data and learning data to form a subspace. Next the residual vector headed from the subspace to the observed data is tracked to identify the anomaly category. To detect anomaly in spite of instability of transient sensing data, the starting point variability of the residual vector is also used to absorb the instability. Experimental results demonstrate that proposed method is beneficial to anomaly detection tasks.