Competitive signature extraction in event forecasting WSNs

Abstract

Signature extraction constitutes an imperative part of a reliable event forecasting system in distributed environments like wireless sensor networks. Recently we published an event forecasting framework which heavily relied on clear, artifact-free event signatures. In this paper we introduce a competitive signature extraction scheme, which can fulfill the criteria needed for reliable event forecasting. Our scheme can continuously keep the events signature database low on artifacts, it can dynamically estimate the number of sequences, and by doing so it is able to continuously extract the event signatures from noisy, overlapped events detected by different sensors in a distributed environment, where the information for a reliable forecast is scattered among the measurements. The method is based on unsupervised (Heb-bian) competitive learning used in self-organizing Kohonen maps. We evaluate the proposed solution by means of simulations and investigate its parameter sensitivity as well.