University of Twente Student Theses


Spatial-temporal outlier and event detection in wireless sensor networks

Amidi, Ali (2013) Spatial-temporal outlier and event detection in wireless sensor networks.

[img] PDF
Abstract:Many spatial phenomena are changing continuously in time and space. Thus, it is the emergence of accessing frequent, up-to-date, spatially dense measurements for monitoring and tracking. Compared to conventional earth observation data collection technologies, wireless sensor networks (WSNs) can provide continuous observations of physical phenomenon by means of dense deployment of sensor nodes. Most WSN applications require accurate, energy friendly and real-time data analysis in order to provide timely information for decision makers. Quality of data provided by WSNs is highly critical while, provided raw data may be drawn in from of a low quality and reliability level expectedly, because of the sensors inexpensive nature. To assure quality of data obtained from WSNs, outlier detection refines the measured data and leads to the retrieval potentially useful information, called an event. There is a trade-off between accuracy and energy efficiency to design the effective event detection method. Temporal and spatial properties of occurred events are important, and thus should receive more attentive consideration. To detect temporal outliers in the absence of a sufficient amount of historical data or in the temporary deployments, building the temporal correlation model is not usually possible thus a novel approach was proposed. The climatic observations and forecasts are utilized to identify outliers. This research evaluated the suitability of using climatic observations and climatic forecasts in WSN for outlier detection. To identify temporal outliers, patterns were utilized instead of absolute values. Experiments from the real dataset from Grand-St.-Bernard deployment in Switzerland and Italy revealed that depending on the degree of localization, forecast values can be directly applied in WSN outlier detection procedures and climatic observations can be used in building the initial temporal model. This method has the advantage of using contextual information in WSNs and was performed in an energy efficient manner. Events were detected based on the correlation based neighbor-voting approach by integrating the temporal and spatial properties of WSNs. Events were identified where a majority of geographically correlated sensors represent temporal outliers, simultaneously. The retrieval of detailed information about spatial and temporal properties of events required the event time scale be degraded and events reclassified as faults and events in finer resolution. Keywords Wireless sensor networks, Temporal outlier detection, Spatial outlier detection, Event detection, MeteoSwiss forecast, MeteoSwiss observation, Patterns
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page