An Energy-Efficient Data Collection Scheme for Wireless Sensor Networks

Keeping data collecting while preserving the scarce energy of sensor nodes is always one of the most crucial problems in wireless sensor networks. In this paper, we propose the DCS scheme to effectively exploit the Ubiquitous temporal-spatial correlation in most natural phenomena for energy-efficient data collection of wireless sensor networks. Specifically, for temporal correlation, we build lightweight AR model locally to capture data distribution at sensor node; for spatial correlation, by making use of our novel definition of similarity measure between sensor nodes, we perform centralized model clustering, which is a kind of clustering that emphasizes data similarity between nodes but ignores geographical distance, to group sensor nodes with similar data distribution on both magnitude and trend into the same cluster. Then through scheduling sensor nodes to report readings alternately and performing dual-prediction at both sensor nodes and Sink, DCS acquires sensing readings without compromising too much data accuracy loss. Simulation results illustrate the efficiency of DCS scheme on a data set synthesized from real-world temperature data, i.e., 82.94% communication overhead reduction while keeping data error as low as 0.0456 degrees C when user-provided error-tolerance threshold sets as 0.2 degrees C.