Transcription
Development of a Framework for DataIntegration, Assimilation, and Learning forGeological Carbon Sequestration(DIAL-GCS)Project #: DE-FE0026515Dr. Alex Sun, PhD, PEBureau of Economic GeologyUniversity of Texas at AustinU.S. Department of EnergyNational Energy Technology LaboratoryMastering the Subsurface Through Technology, Innovation and Collaboration:Carbon Storage and Oil and Natural Gas Technologies Review MeetingAugust 16-18, 2016
Acknowledgements DOE/NETL: Brian Dressel UT– BEG: Sue Hovorka (geologist) Hoonyoung Jeong (petroleum engineer)Katherine Romanak (geochemistry)Sergey Fomel, Tieyuan Zhu (geophysics)Clay Templeton (web development)– Information School: David Arctur (data scientist)– Texas Advanced Computing Center: Weijia Xu (HPC)2
Outline Benefits to the program Technical status Accomplishments to date Synergy opportunities Summary3
Benefit to the ProgramCarbon storage program goals being addressedDevelop and validate technologies to ensure 99 percent storagepermanenceIMS: integrative carbon storage reservoir management technology thatcombines real-time measurement of reservoir properties with projectspecific data management and data processing workflowsExpected benefits of this IMS ProjectTransform scientific knowledge to decision knowledge and publicknowledge: Promote data sharing and visual analytics– Better collaboration among team members– Public outreach Streamline GCS management and decisionmaking Facilitate the optimal allocation of monitoring resources4
Challenges and Motivations GCS is highly interdisciplinary Data sharing and discovery is difficult Lack of project-specific risk assessment and dataassimilation tools Lack of systematic demonstration of machine learningtechnologies5
Project Overview:Goals and ObjectivesA. Develop GCS data management module for storing,querying, exchanging, and visualizing GCS data frommultiple sources and in heterogeneous formats–Success Criterion: Whether a flexible, user-friendly Web portalis set up for enabling data exchange and visual analyticsB. Incorporate a complex event processing (CEP) enginefor detecting abnormal situations by seamlesslycombining expert knowledge, rule-based reasoning, andmachine learning–Success Criterion: Whether a set of decision rules aredeveloped for identifying abnormal signals in monitoring data6
Project Overview:Goals and ObjectivesC. Enable uncertainty quantification and predictive analyticsusing a combination of coupled-process modeling, dataassimilation, and reduced-order modeling– Success Criterion: Whether a suite of computational tools aredeveloped for UQ and predictive analyticsD. Integrate and demonstrate the system’s capabilities withboth real and simulated data– Success Criterion: Whether the IMS tools developed underGoals A to C are integrated, streamlined, and demonstrated for arealistic GCS site7
Technical Status8
System Architecture of DIAL-GCSTask 2: Sensor Data Schema Development and Serialization (Y1)Task 3: Development of CEP, machine learning (Y1-2)Task 4: Coupled Modeling and Data Assimilation (Y1-3)Task 5: Integration and Demonstration (Y1-3)9
Task 2: Sensor Data Management10
DIAL Frontend11
Web-GIS Admin12
A Dashboard for Visual Analytics Live monitoring datastreamsConnect to different NoSql serversWeb-based database query and visualization interface13
Mobile Frontend14
Task 3: CEP DevelopmentThree event processing stages Signaling– Detection of abnormal event Triggering– Associate an event with a pre-defined rule orrule set Evaluation– Provide diagnostics/reasoning on an event15
Anomaly DetectionSensing TechnologyCategoryData(temporal/spatial)Possible EventSeismic imagingGeophysicalDiscrete/vectorTime-lapse changeDownhole pressuregaugeGeophysicalContinuous/scalarAbrupt change from anominal trendSoil gas samplingGeochemicalDiscrete/scalarChange in gas ratio (CO2 vs.O2)Groundwater samplingGeochemicalDiscrete/scalarSampled values (pH,dissolved CO2) exceedingthresholdContinuous/vectorTemperature changeDistributed temperature Geophysicalsensing16
Demo Dataset17
Detection of LeakageRaw DataDownhole Pressure CPTLevel 0: Raw DataLevel 1: EventsLevel 2: Aggregated events and alarms18
Flow Meter DataRaw DataFlow Meter CPTLevel 0: Raw DataLevel 1: EventsLevel 2: Aggregated events and alarms19
Web-Based Anomaly Detection(Preliminary)20
Beyond Time/Scalar TuplesLeak EventsData Courtesy: Barry Freifeld21
Task 4 Modeling & Data Assimilation Phase 1: Develop “virtual observatories” atfor testing monitoring design, riskmanagement, uncertainty quantification– All models need to make use of highperformance computing– Main focus: Flow, Seismic, and Geochemistry Phase 2: Integrate process-level modelswith anomaly detection and dataassimilation22
General Workflow Perform ensemblesimulation on TACC Send results toserverModelingAnomalyDetection Compare predictiveanalytics toobservations Learning: Detectanomalies Assimilation: Usedata to improvemonitoring design Ensemble KalmanfilterDA23
Flow ModelingEnsembleOf RealizationsUQ, Predictive Analytics, and Machine LearningPerform ensemblesimulationAnomaly detectionBy Hoonyoung Jeong 24
Seismic-Based Anomaly Detection Role:– Virtual observatory for assessing efficacy ofseismic surveys for picking up leakage– Quantify time-lapse response of seismicvelocity to different leak scenarios or features Methodology– Create 3d velocity and density models usingwell logs and reflection seismic dataBy Tieyuan Zhu & Sergey Fomel25
Two velocity models (in time)Pre-injectionPost-injectionafter 1 year26
2D modeling: inline velocity slicesPre-injectionPost-injectionafter 12 monthThe two slices arewindowed from theprevious twovelocity models.27
Simulated seismic data: 2D slicesPre-injectionPost-injectionafter 1 yearDifference28
Geochemical Data Inversion Roles– Ingesting geochemical data for anomalydetection– Leak source attribute recovery Methodology– Adopted a flexible parameterization technique(level set) for representing source geometry29
Shallow Aquifer Leak Source RecoveryForward model: TOUGHREACT30
Accomplishments to DateDescriptionTaskDeveloped prototypes for demonstrating datastream management,visualization, and Web-GIS, including capabilities to load data fromdifferent sensor sources;2, 5Developed multiphysics models, including a multiphase CO2 simulationmodel, a forward seismic time-lapse response model, and geochemicaldata inversion capabilities4Implemented anomaly detection algorithms and integrated with the Web 331
Summary & Future Work Year 1 activities focus on IMS infrastructure building Developed Web-based protocol for demonstrating allaspects of IMS Our current IMS has user-friendly dashboard and AIcapabilities Developed modeling capabilities In Year 2, we will continue to Enhance data visualization capabilityImplement more complex event processing algorithmsMultiphysics model integrationDevelop monitoring network design capabilitiesBuild web-based UQ and predictive analysis capabilities32
Synergy Opportunities– EDX: Big Data computing for data discovery andanomaly detection– NRAP: reduced models– Collaborate with other teams to demonstrateIMS capabilities33
Appendix– These slides will not be discussed during thepresentation, but are mandatory34
Organization ChartYoungBEGAssociate DirectorSunHovorkaRomanak(Co-PI)Yang(PI)Xu (TACC)FomelGradudateStudentsArctur (UTInformation School)35
Gantt Chart36
Bibliography Sun, A., Islam, A., Wheeler, M., Identifying Attributes of CO2 LeakageZones in Shallow Aquifers Using a Parametric Level Set Method,submitted to Greenhouse Gases: Science and Technology Sun et al., Development of an Intelligent Monitoring System forGeological Carbon Sequestration (GCS). Abstract submitted to 2016Fall Meeting of American Geophysical Union.37
using a combination of coupled-process modeling, data assimilation, and reduced-order modeling - Success Criterion: Whether a suite of computational tools are developed for UQ and predictive analytics D. Integrate and demonstrate the system's capabilities with both real and simulated data
