Physical clustering of nodes in sensor networks aims at grouping together sensor nodes according to some similarity criteria like neighborhood. Out of each group, one selected node will be the group representative for forwarding the data collected by its group. This considerably reduces the total energy consumption, as only representatives need to communicate with distant data sink. In data mining, one is interested in constructing these physical clusters according to similar measurements of sensor nodes. Previous data mining approaches for physical clustering concentrated on the similarity over all dimensions of measurements. We propose ECLUN, an energy aware method for physical clustering of sensor nodes based on both spatial and measurements similarities. Our approach uses a novel method for constructing physical clusters according to similarities over some dimensions of the measured data. In an unsupervised way, our method maintains physical clusters and detects outliers. Through extensive experiments on synthetic and real world data sets, we show that our approach outperforms a competing state-of-the-art technique in both the amount of consumed energy and the eectiveness of detecting changes in the sensor network. Thus, we achieve an overall signicantly better life times of sensor networks, while still following changes of observed phenomena.
Self-Organizing Energy Aware Clustering of Nodes in Sensor Networks Using Relevant Attributes
Palpanas, Themistoklis;
2010-01-01
Abstract
Physical clustering of nodes in sensor networks aims at grouping together sensor nodes according to some similarity criteria like neighborhood. Out of each group, one selected node will be the group representative for forwarding the data collected by its group. This considerably reduces the total energy consumption, as only representatives need to communicate with distant data sink. In data mining, one is interested in constructing these physical clusters according to similar measurements of sensor nodes. Previous data mining approaches for physical clustering concentrated on the similarity over all dimensions of measurements. We propose ECLUN, an energy aware method for physical clustering of sensor nodes based on both spatial and measurements similarities. Our approach uses a novel method for constructing physical clusters according to similarities over some dimensions of the measured data. In an unsupervised way, our method maintains physical clusters and detects outliers. Through extensive experiments on synthetic and real world data sets, we show that our approach outperforms a competing state-of-the-art technique in both the amount of consumed energy and the eectiveness of detecting changes in the sensor network. Thus, we achieve an overall signicantly better life times of sensor networks, while still following changes of observed phenomena.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione