Agent Centric Sensor Network Association using Similarity Measures
R Vasquez Stacey, Martin Colley · 2008
Inside the grouping process of sensor networks each node must decide what local group it is going to be a part of for data aggregation and dissemination. We look at how to form the most likely groups using agent centric methods based on the similarity to other nodes in the network and evaluate methods based on clustering, thresholding and fuzzy logic. The methods use simple scores that represent the similarity to local nodes and are optimised using a genetic algorithm in simulation. Using these methods we achieve an accuracy of around 80% in simulation of a large number of nodes using data obtained from real world data-logging. These results are validated using real world experimentation and we show that fuzzy thresholding outperforms the other methods.