Spatio-Temporal Compressive Sensing Technique for Data Gathering and Anomaly Detection in Wireless Sensor Networks

In this paper, we focus on collecting data and detecting anomalies in Wireless Sensor Networks (WSNs) while optimizing the use of sensor computational and energetic resources. Recently, Compressive Sensing (CS)-based solutions had been the subject of extensive studies for the design of efficient data gathering solutions in WSNs. However, existing CSbased approaches are very sensitive to outlying values and do not offer a proper tool to deal with the presence of anomalies. Moreover, CS data gathering schemes are based only on the spatial correlation pattern between sensory data and ignore an important feature which is the temporal correlation between sensor readings. This paper introduces a novel CS-based data gathering solution that allows to integrate the spatial and temporal correlation features into the data recovering process. Furthermore, the proposed approach is built in such a way that it also allows to detect and correct eventual anomalies. We propose a general formulation of data gathering and anomaly detection problem as a tractable convex optimization problem on the Hilbert space of data measurements and anomalies. Besides, we design a new class of primal-dual algorithms to solve the resulting optimization problem. We evaluate the efficiency of our method by running extensive simulations on two real datasets. We demonstrate that the proposed algorithm achieves good data recovery and anomaly detection performance and outperforms the main state-of-the-art technique addressing the same problem.