Poster: Towards an Inexact Semantic Complex EventProcessing FrameworkQunzhi ZhouDepartment of ComputerScienceUniversity of [email protected] SimmhanMing Hsieh Department ofElectrical EngineeringUniversity of [email protected] PrasannaMing Hsieh Department ofElectrical EngineeringUniversity of [email protected] event processing (CEP) deals with detecting real-time situations, represented as event patterns, from amongan event cloud. The state-of-the-art CEP systems processevents as plain data tuples and are limited to detect pre-cisely defined patterns. Emerging application areas like op-timization in smart power grids require CEP to incorporatesemantic knowledge of the domain for easier pattern speci-fication, and detect inexact patterns in the presence of un-certainties. In this paper, we present motivating use cases,discuss limitations of existing CEP systems and describe ourwork towards an Inexact Semantic Complex Event Process-ing (InSCEP) framework.Categories and Subject DescriptorsH.4 [Information Systems Applications]: MiscellaneousGeneral TermsDesign, LanguagesKeywordsComplex event processing, Semantic Web, demand response1. INTRODUCTIONComplex event processing deals with detecting real-timesituations, represented as event patterns, from among anevent cloud. In recent years, research into CEP has receivedmuch attention in the research community motivated by ap-plications in domains like financial services [2] and RFIDdata management [4]. Many research prototypes, and sev-eral commercial systems such as ruleCore1and Esper2havebeen developed.Demand response optimization (DR) in Smart Grid is anemerging application area for CEP [5]. Smart Grid is themodernization of power grid by integrating digital and in-formation technologies, with the deployment of millions ofsensors and smart meters to monitor energy use activities.1http://www.rulecore.com2http://esper.codehaus.orgCopyright is held by the author/owner(s).DEBS’11, July 11–15, 2011, New York, New York, USA.ACM 978-1-4503-0423-8/11/07.DR, a cornerstone application of Smart Grid, deals with cur-tailing power load when a peak load is encountered. Con-tinuous data relevant to DR emanating from various sourcescan be abstracted as events. These may be from smartappliances (ThermostatChange event), smart meters (Me-terUpdate event), weather phenomena (HeatWave event) orconsumer activity (ClassSchedule event). CEP can correlatethese heterogeneous events to detect patterns that predictpeak load occurrences or identify load curtailment opportu-nities for DR in a timely manner.Limitations of existing CEP systems limit their uses indiverse information space like Smart Grid. Existing systemsprocess events as relational data tuples. As such, event pat-terns can only be defined as a combination of attributes pre-sented in event data. In addition, most CEP systems onlysupport precise pattern matching, without any leeway torelax pattern constraints. However, uncertainty is an intrin-sic feature of real world cyber-physical applications, wherepotentially incomplete and even incorrect information exist,yet need to be matched within certain bounds.An effective CEP solution for DR optimization needs toextend traditional CEP systems in two aspects. First, itmust be extensible to meet the organic growth of the SmartGrid information diversity with the provision to easily modeland identify new events and event patterns by both domainexperts and non-domain users. Second, it should captureuncertainties of events, and relax deterministic event pat-terns for inexact pattern detection.2. USE CASESWe present example DR event patterns for load predic-tion, curtailment and monitoring, and use them to illus-trate key features that our proposed Inexact Semantic Com-plex Event Processing (InSCEP) framework should provide.Consider in a campus micro grid, the DR application pro-cesses information coming from sensors and equipments thatreport their measurements or operations. We have the fol-lowing patterns,i. Load Prediction: A teaching building consumes 90% ofits peak load, more than 5 classrooms have high proba-bilities of increasing from base load according to meterreadings, class schedules and weather conditions.ii. Load Curtailment: The thermostat in one office roomis tuned 5 degrees lower than the average set point ofthermostats in the same type of rooms which were tunedin the last 30 minutes.iii. Load Monitoring: Conservative curtailment patterns wereapplied, followed by a sequence of meter readings thatindicate power load remains steady or increases.Traditional CEP systems define patterns by specifyingprecise constraints of event data. However, the above exam-ples illustrate the need to incorporate semantics and flexi-bility in pattern specification. The background knowledgeof events from multiple domains (e.g. electrical systems,appliances, room scheduling, etc) need to be captured. Inaddition, flexibility has to exist to allow a limited number oferrors or mismatches to still detect a relevant pattern. Theneed for specify such inexact patterns lies in two reasons, (1)component events can be probabilistic due to imprecise orincomplete observations, and (2) event pattern itself is un-certain and may have infinite acceptable equivalences. Forinstance, in the third example the sequence of meter readingsneed not strictly remain constant or monotonically increase.A small fraction of outsider readings should be tolerated.3. INSCEP: SYSTEM DESCRIPTIONFigure 1: CEP (top) and InSCEP (bottom)The proposed InSCEP framework for demand responseoptimization extends traditional CEP with Semantic Webtechnologies and uncertain pattern detection. As shown inFigure 1, our approach allows users to specify inexact eventpatterns based on a domain-specific event model at highlevel. As the system processes events, it annotates sourcedata using the semantic event model, correlates events toderive unobserved events and then performs inexact and se-mantic matching to detect pattern occurrences.Semantic and Uncertain Event ModelWe develop an event model that captures semantics anduncertainties. Events are modeled using Semantic Web on-tologies and represented using the Web Ontology Language(OWL). The event ontology is modular, for easy extensionwith new domain ontologies. The top level, core event on-tology captures concepts and relationships between events.The notion of an event is classified into