From Real Time Sensor Streams to Real Time Event Streams Harshal Patni Kno.e.sis Center Wright State University Dayton, Ohio, USA 1-937-775-5215 harshal@knoesis.org Cory Henson Kno.e.sis Center Wright State University Dayton, Ohio, USA 1-937-775-5215 cory@knoesis.org Amit Sheth Kno.e.sis Center Wright State University Dayton, Ohio, USA 1-937-775-5215 amit@knoesis.org Pramod Anantharam Kno.e.sis Center Wright State University Dayton, Ohio, USA 1-937-775-5215 pramod@knoesis.org ABSTRACT Sensors are increasingly being deployed for continuous monitoring of physical phenomena, resulting in avalanche of sensor data. Current sensor data streams provide summaries (e.g., min., max., avg.) of how phenomena change over time; however, such summaries are of little value to decision makers attempting to attain an insight or an intuitive awareness of the situation. Event-streams, on the other hand, provide a higher- level of abstraction over the sensor data and provide actionable knowledge useful to the decision maker. In this work, we present an approach to generate event-streams in real-time. This is accomplished through the application of ontological domain knowledge in order to integrate sensor data streams and infer the existence of real-world events. The generated event-streams are publicly accessible on the Linked Open Data (LOD) Cloud. Categories and Subject Descriptors C.2.1 [Sensor Networks] General Terms Design. Keywords Event Streams, Sensor Data Streams, Linked Open Data Cloud (LOD), LinkedSensorData, Semantic Sensor Web, Sensor Web Enablement, Resource Description Framework (RDF), SPARQL, Semantic Web, Ontologies 1. INTRODUCTION Sensors produce huge amounts of data about our environment that arrives rapidly in continuous and time-varying streams [1]. For example, a sensor system on a Boeing 737 freighter generates data at the rate of up to 11.37 GB per second for a six hour flight. This data stream would quickly overwhelm any system not capable of effectively detecting and analyzing the most important data (see [2]). Analyzing such sensor data streams and providing meaningful abstractions in real-time presents a significant research challenge. Weather data aggregation services like MesoWest[3] provide summarizations (such as minimum, maximum, and average values (see [4])) of sensor data streams over a period of time. These summarizations are necessary, but not sufficient, for presenting meaningful abstractions that humans can easily comprehend and utilize for effective decision-making. For example, such summarizations cannot help in answering questions that involve real world events, such as: Which weather stations are currently detecting a Blizzard? or, What event (or sequence of events) is currently being detected by a weather station? The ability to answer these questions requires the integration of data from multiple sensor streams. In addition, semantic analysis must be performed, using external domain knowledge to infer higher-level abstractions, or “events”. An event-stream can be generated by aggregating a sequence of event abstractions for a particular sensor, over a period of time. Event-streams provide a clear and intuitive representation of how events evolve over time. An intuitive representation of trends in events will present decision makers with actionable situation awareness; and making the event- streams available on LOD provides global accessibility. 2. Background The concept and capabilities for generating real-time event streams from real-time sensor streams is part of a comprehensive project on Semantic Sensor Web (SSW) 1 [5] and can be seen as one example of a broader vision of EventWeb [6] outlined by Ramesh Jain. SSW adapts a semantic Web approach to making heterogeneous, multimodal sensor data more meaningful. This is accomplished by automatic annotation of sensor data (i.e., metadata) and the use of semantic query processing and reasoning for advanced processing in order to support semantic search, integration, mining, analysis and situational awareness over sensor data. Annotation is usually done with respect to a variety of ontologies. Ontologies such as the sensor and observation ontology are used to support interoperability over heterogeneous sensors. Domain specific ontologies pertaining to domains such as weather and environment are also used. Within the sensors community, there is a broad acceptance and adoption of Sensor Web Enablement (SWE) 2 standards – developed by the Open Geospatial Consortium (OGC) 3 – and utilization of service oriented architecture to support system and syntactic interoperability. Consequently, a number of SSW realizations – including ours – extend SWE with Semantic Web capabilities. A number of demonstration systems have been developed as part of our SSW project including Sensor Discovery on Linked Data and Semantic Sensor Observation Service (SemSOS), 1 http://semantic-sensor-web.com/ 2 http://www.opengeospatial.org/ogc/markets-technologies/swe 3 http://www.opengeospatial.org/