Predicting Cluster Formation in Decentralized Sensor Grids Astrid Zeman and Mikhail Prokopenko CSIRO Information and Communication Technology Centre Locked bag 17, North Ryde, NSW 1670, Australia mikhail.prokopenko@csiro.au Abstract. This paper investigates cluster formation in decentralized sensor grids and focusses on predicting when the cluster formation converges to a stable con- figuration. The traffic volume of inter-agent communications is used, as the un- derlying time series, to construct a predictor of the convergence time. The predic- tor is based on the assumption that decentralized cluster formation creates multi- agent chaotic dynamics in the communication space, and estimates irregularity of the communication-volume time series during an initial transient interval. The new predictor, based on the auto-correlation function, is contrasted with the pre- dictor based on the correlation entropy (generalized entropy rate). In terms of predictive power, the auto-correlation function is observed to outperform and be less sensitive to noise in the communication space than the correlation entropy. In addition, the preference of the auto-correlation function over the correlation entropy is found to depend on the synchronous message monitoring method. 1 Introduction There is a distinction between “Sensor Networks” and “Sensor Grids”, as pointed out in recent literature (e.g., [3]): “whereas the design of a sensor network addresses the logical and physical connectivity of the sensors, the focus of constructing a sensor grid is on the issues relating to the data management, computation management, informa- tion management and knowledge discovery management associated with the sensors and the data they generate”. One significant issue addressed by sensor grids is dynamic sensor-data clustering, aimed at grouping entities with similar characteristics together so that main trends or unusual patterns may be discovered. This is investigated as decen- tralized clustering in multi-agent Systems [9], dynamic cluster formation in mobile ad hoc networks [7] and decentralized sensor arrays [8, 13, 10]. The latter studies describe dynamic cluster formation as self-organisation of dynamic hierarchies, with multiple cluster-heads emerging as a result of inter-agent communications, and indicates that decentralized clustering algorithms deployed in multi-agent systems are “hard to eval- uate precisely for the reason of the diminished predictability brought about by self- organisation”. The results presented in [13] identified a predictor for the convergence time of dynamic cluster formation, based on the traffic volume of asynchronous inter- agent communications. Following this study, we attempt to adapt a decentralized clus- tering algorithm to a specific topology (a rectilinear grid) and replace a complicated predictor with a more simple measure, based on synchronized aggregation of multi- agent communications.