| Summary: | Wireless sensor networks have been widely deployed for environment monitoring. The resource-limited sensor nodes usually transmit the sensing readings to Sink node collaboratively in a multihop manner to conserve energy. In this paper, we consider the problem of spatial clustering for approximate data collection that is feasible and energy-efficient for environment monitoring applications. Spatial clustering aims to group the highly correlated sensor nodes into the same cluster for rotatively reporting representative data later. Through a thorough investigation of a real-world environmental data set, we observe strong temporal-spatial correlation and define a novel similarity measure metric to inspect the similarity between any two sensor nodes, which take both magnitude and trend of their sensing readings into consideration. With such metric, we propose a clustering algorithm named as HSC to group the most similar sensor nodes in a distributed way. HSC runs on a prebuilt data collection tree, and thus gets rid of some extra requirements such as global network topology information and rigorous time synchronization. Extensive simulations based on realworld and synthetic data sets demonstrate that HSC performs superiorly in clustering quality when compared with the alternative algorithms. Furthermore, approximate data collection scheme combined with HSC can reduce much more communication overhead while incurring modest data error than with other algorithms.
|
|---|