Article Preview
Top1. Introduction
Wireless Sensor Networks (WSNs) consist of distributed wirelessly connected sensors capable of sensing (observing) specific phenomena. WSNs are adopted to support novel applications that efficiently respond to any abnormalities in sensors readings. Each sensor has specific sensing abilities for performing measurements related to a phenomenon under consideration. The most important advantage of such a setting is the autonomous nature of sensors. When deployed in a field of interest (e.g., a forest, an energy network), they can automatically perform measurements and disseminate them to their spatial neighbors until reaching a central processing system, hereinafter referred to as Central System (CS). During the past years, a large set of critical applications have been developed on top of streaming contextual data. A CS collects contextual data and processes them to (i) identify certain phenomena and then (ii) react to specific events. These events are related to critical aspects such as security issues or violations of pre-defined constraints.
WSNs are widely adopted in monitoring applications in various domains. In security applications, a monitoring infrastructure is imperative by adopting a CS which applies a fast and efficient mechanism to derive alerts when specific criteria are met (Kausar et al., 2012; Rothenpelier et al., 2009). Such criteria are related to sensor failures, resources depletion, or other abnormalities. It is of paramount importance that the identification of failures should be in (near) real time as time-critical applications require immediate responses to eliminate any negative consequences. Another interesting application domain is environmental monitoring (Bourgeois et al., 2003; Gouveia & Fonseca, 2008; Hardas et al., 2008). Environmental monitoring has attracted significant interest as any negative effect in the environment heavily affects human lives. The key aspect of a CS is to be pro-active and immediately respond to any change in the environment. Many research and commercial CSs adopt (i) sensors observing a specific phenomenon (e.g., temperature, humidity, water level, pollution) and (ii) an intelligent mechanism that responds to the identification of events (e.g., fire, flood). In addition, monitoring applications over WSNs could offer many advantages in energy management (Kumar, 2011; Kesav & Rahim, 2012; Thamarai & Amudhevalli, 2014). A set of sensor could undertake the responsibility of monitoring the power consumption and act in a pro-active manner in order to timely detect abnormalities and support the appropriate actions for securing the energy delivery to consumers or industry.
Information and Communication Technologies (ICT) could offer many advantages in contextual streams monitoring. Machines could undertake the responsibility of the continuous monitoring process and result the appropriate actions in any observed abnormalities. A number of sensors could monitor specific areas or systems and the CS could reason over the observed values and derive decisions related to the appropriate responses for every abnormality. In this paper, we propose a mechanism that combines data fusion techniques, prediction (time series regression) and Fuzzy Logic (FL) to derive a decision making tool for the identification of events. The proposed mechanism builds on top of measurements reported by a number of sensors and aims to provide immediate responses to any observed abnormality. Our mechanism derives knowledge from the team of sensors and does not rely on single sensor observations. A single sensor could be affected by a number of reasons (e.g., location where the sensor is placed, network connection, battery level) and its reports could not be valid. The proposed system aggregates the reported measurements and reasons over the opinion of the team about the identification of the event. The aim is to efficiently handle the event, however, to minimize false alarms / alerts that could affect the performance of the response. The contribution of our paper is as follows: