Abstract:
Sensor networks used in environmental monitoring applications are subject to harsh environmental conditions and hence are prone to experience failures in its measurements. Comparing to the common task of outlier detection in sensor data, we review herein the complex problem of detecting systematic failures such as drifts and offsets. Performing this detection in environmental monitoring networks becomes a stringent task especially when we need to distinguish data errors from real data deviations due to natural phenomenon. In this paper, we detail the scope of events and failures in sensor networks and, considering those differences, we introduce a new instantiation of a proven methodology for dependable runtime detection of outliers in environmental monitoring systems to address drifts and offsets. Lastly, we discuss the use of machine learning techniques to estimate the network sensors measurements based on the knowledge of processed past measurements alongside with the current neighbor sensors observations.
Keywords:
Data quality, Failure detection, Sensor fusion, Machine learning, Sensor networks, Aquatic monitoring.
_______________________________________________________________________
