Data collection in wireless sensor networks (WSNs) can become extremely expensive in terms of power consumption if all measurements have to be fetched. However, since multiple applications do not require data from all nodes but to compute a function over a smaller data set, much of the available data on the network can be considered irrelevant and not worthy of spending energy. In this context, in-network filtering schemes can be used to forward only relevant data towards a sink node for processing purposes. In this work, we propose and evaluate two schemes that can drive this filtering process. Both of them are based on the integration of metaheuristics and learning algorithms inspired by nature. In particular, we consider the computation of the maximum function as case study for these schemes. We investigate the trade-off between communications costs, which are directly associated with power consumption, and error costs due to fetching not all relevant data. We show by simulation that communication costs can be significantly reduced with respect to traditional schemes while keeping the computation error bounded.
展开▼
