Improving data aggregation in Wireless Sensor Networks with time series estimation

Abstract

Wireless sensor networks (WSNs) are widely deployed nowadays on a large variety of applications. The major goal of a WSN is to collect information about a set of phenomena. Such a process is non-trivial since batteries' life is limited and thus wireless transmissions as well as computing operations must be minimized. A common task in WSNs is to estimate the sensed data and to spread the estimated samples over the network. Thus, time series estimation mechanisms are vital for this type of processes so as to reduce data transmission. In this paper, we assume a single-hop clustering mechanism in which sensor nodes are grouped into clusters and communicate with a sink through a single hop. We compare three mechanisms based on the autoregressive, moving average, and normalized least mean square models to predict local sensed samples in order to reduce wireless data communication. We prove the efficiency of these algorithms with real samples publicly available and show how suitable they are for processes of different nature.