Event detection systems can help detecting incidents of interest that occur within data streams generated by interconnected sensor devices. Corresponding personnel need to be notified about the incidents so that appropriate actions can be taken to address them. One of the typical use cases of event detection systems is monitoring the quality of recycling water during wastewater treatment. Existing event detection systems in this domain usually focus on dealing with low level primitive events and thus are not capable of detecting more complex events which contain relations and background knowledge. In this paper, we present a framework for detecting complex events with background knowledge in event streams, by using ontology and query rewriting techniques.
In today’s business environment, complex events occurs in real time and companies can benefit from efficient real-time processing of data streams generated by interconnected sensor devices to discover such complex events. However, some complex events cannot be described using low-level event detection rules based on pattern-matching. The definition of such events usually requires domain expertise and background knowledge. Using ontology and rules to model complex events significantly enhances the accuracy of detecting such events. One of the use cases is monitoring the quality of recycled wastewater, where the water quality is monitored using sensing devices. A system that is capable of detecting complex events can provide an early indication of water quality changes, for relevant personnel to take appropriate actions to adjust the water treatment accordingly.
Existing stream modeling ontologies such as the SSN/SOSA [
In this paper, we present a framework for detecting complex events with rich background knowledge over event streams, by using ontology and query rewriting techniques. We evaluate the performance of our framework and event detection using historical time-series data collected from one of the wastewater treatment plant in Australia and queries representing common water-quality-change events.
Our framework contains three components highlighted in blue in Figure 1. Sensor
data flow into the framework as streams, and the ontology component validates and
transforms the data into a format which streaming component can process. Users
specify the events of interest as a set of queries Q. The query rewriting component
rewrites Q into a more complicated query Q ’ based on the ontology model, and then
delivers the transformed query Q ’ to streaming component. Finally, the transformed
data run through Q ’ at the streaming component and the events of interest are
detected through query answering. The framework was designed to be modular, so
that each component can cater for different technologies.
axioms to create more specific queries from the original most general query [
For example, users may want to ask ‘What time does an event of interest occur at which barrier of the treatment plant A?’ This question is translated to the below SPARQL query: SELECT ?t ?b ?s WHERE {
?s a Barrier; name ?b; hasEvent [a ComplexWaterEvent] .
FILTER (?s hasName ‘A’)} Axiom: IndustryDischarge subClassOf ComplexWaterEvent After query rewriting based on the ontology axiom (i), query (1) is transformed into SELECT ?t ?b ?s WHERE { {?s a Barrier; name ?b; hasEvent [aComplexWaterEvent]} UNION {?s a Barrier; name ?b; hasEvent [a IndustryDischarge]} FILTER(?s hasName ‘A’)} The component was designed to be modular so that any query rewriting tools can be used in the component. The original queries Q are transformed into Q ’ after rewriting, and then propagated to the streaming component.
The streaming component is the place where incoming streams of the transformed data meets the rewritten query Q ’. There are several candidates which are suitable to use as our streaming component, such as Apache Spark, Storm etc. An alert is generated as output of this component, once an event of interest is detected.
We tested our framework in the wastewater domain, using historical time series data collected from one of the wastewater treatment plant in Australia. The original data was in CSV format which consists total number of 79306 individual records. We first simulated the streaming data using a small program which reads the CSV file and ingest each record at speed of 30 seconds interval per record. Then, we added 2 extra classes which defines 3 types of complex events, ‘Industry Discharge’, ‘Heavy Rainfall’ and ‘Plant Fault’ in the wastewater domain to our original ontology. These 3 complex events are our targets to be discovered during the evaluation of our framework. The ontology we used in this test is depicted in Figure 2.
For the query rewriting component, we adopted Ontop [
In this paper, we presented an ontology based framework to detect meaningful event and evaluated it using data collected from wastewater treatment process. Our system provides a innovative way for event detection in the water monitoring domain. In our future work, we would be focusing on reducing the system raw data conversion time, as well as enhancing the ability of detect multiple and more complex events. ¹ Although the latest release notes claim that Ontop has been incorporated, but it seems query rewriting has not been fully implemented.