jpathak/dddas/index.html Session … · 5 Basis for project: a dynamic data driven application...
Transcript of jpathak/dddas/index.html Session … · 5 Basis for project: a dynamic data driven application...
11
Integrated Decision Algorithms for Auto-steeredElectric Transmission System Asset Management
JimJim McCalley, Yuan Li: McCalley, Yuan Li: ElectricalElectrical EngineeringEngineering –– PowerPower SystemsSystems
Vasant Honavar, J. Pathak: Comp. Vasant Honavar, J. Pathak: Comp. ScienceScience –– Data Data IntegrationIntegration, Software , Software designdesign
Bill Meeker, Y. Hong: Statistics Bill Meeker, Y. Hong: Statistics –– Reliability, DecisionReliability, Decision
Daji Qiao, Computer Engineering Daji Qiao, Computer Engineering –– Sensor NetworksSensor Networks
Ron Roberts: Aerospace Engineering Ron Roberts: Aerospace Engineering –– Nondestructive EvaluationNondestructive Evaluation
Sarah Ryan, Ye Mujing:Sarah Ryan, Ye Mujing: Industrial Engineering Industrial Engineering –– Stochastic OptimizationStochastic Optimization
Iowa State UniversityIowa State UniversityNSF NSF AwardAward CNS0540293CNS0540293
www.cs.iastate.edu/~jpathak/dddas/index.htmlSession W42d (10-12 am) on
Dynamic Data-Driven Application Systems ICCS – 2007, Beijing, May 29, 2007
22
OutlineOutline• Project overview
Motivation and basis for project
A 5-layer design: Software architecture
Layer 1: Long-term power system simulation
Layer 2: Sensing and communications
Layer 3: Data integration
Layer 4: Data transformation
Layer 5: Simulation & decision
• Integrated decision problems: Benders, example
33
Motivation: Ages of two US companyMotivation: Ages of two US company’’s s transformerstransformers
PJM: 55% are within 10 years of design life
For this company, 60% are within 15 years of design life
44
Solution: MonitorSolution: Monitor
Serveron Corporation Launched to Monitor the Performance of Critical Electric Power Transmission and Distribution EquipmentSan Diego, Calif. — DistribuTECH 2001 ÷February 5, 2001 — Today at the DistribuTECH2001 show in San Diego, Serveron Corporation announced its launch as the electric power industry's first dedicated provider of turn-key equipment and services to monitor the health of electric generation, transmission, and distribution substation equipment.
Power Monitor Solutions Welcome to Sierra Pacific…Specializing in affordable next generation power monitoring,predictive maintenance and infrared camera solutions.
ABB wins consulting order to increase power equipment reliability in U.S. 2003-09-15 - Asset management services will help track critical equipment Zurich, Switzerland, September 15, 2003 - ABB, the leading power and automation technology group, today announced it has won an order to help International Transmission Company in the United States manage its power transmission assets.
55
Basis for project: a dynamic data driven Basis for project: a dynamic data driven application systemapplication system
••Monitoring technologies proliferating, provide Monitoring technologies proliferating, provide continuous data streaming of equipment state continuous data streaming of equipment state
••Presently used for detection and diagnosisPresently used for detection and diagnosis
••We want to use for failure predictionWe want to use for failure prediction
••Failure prediction is essential for decision: operations, Failure prediction is essential for decision: operations, maintenance, investment, sensor deploymentmaintenance, investment, sensor deployment
Develop HW/SW prototype for autoDevelop HW/SW prototype for auto--steering steering informationinformation--decision cycles inherent to managing decision cycles inherent to managing operations, maintenance, & planning of highoperations, maintenance, & planning of high--voltage voltage electric power transmission systems.electric power transmission systems.
66
Algorithms for system-
level decision
Failure indices
Monitors?
Monitor
Xfmrs, CBs, Lines
Substation communication
box
STAGE 1
STAGE 2(Scenario 1)
x*
y1*
+c(x*)
P1d(y1*)
Min
STAGE 2(Scenario 2)
y2*P2d(y2*)
Min
x*
Expand?
Maintain?
Operate?
Dynamic Data-Driven Asset Management
Data Integration -Transformation
System
Wide area network
Condition Data
Power Grid (can be simulation)can be simulation)
77
Software Design: ServiceSoftware Design: Service--Oriented ArchitectureOriented ArchitecturePower system asset management: Power system asset management: geographically/physically distributed assets geographically/physically distributed assets and processes, heterogeneity in data and processes, heterogeneity in data semantics, interoperability issuessemantics, interoperability issues
need a software design to accommodate need a software design to accommodate integrationintegration
PSAM: Power System Asset ManagementPSAM: Power System Asset Management
ServiceService--oriented architectureoriented architecture–– Web Service designed to support interoperable Web Service designed to support interoperable
machinemachine--22--machine interaction over a networkmachine interaction over a network–– Facilitates handling heterogeneous data semanticsFacilitates handling heterogeneous data semantics–– Has commercial support, based on standards: Has commercial support, based on standards:
SOAP, WSDL, WSRFSOAP, WSDL, WSRF
J. Pathak,Y. Li, V. Honavar, J. McCalley, “A Service-Oriented Architecture for Electric Pwr Transmission System Asset Mngmnt,” 2nd
Intrntnl Wrkshp on EngrService-Oriented Apps: Design & Composition, Dec. 4, 2007, Chicago, Ill.
88
PSAM PSAM FrameworkFramework
Domain Specific OntologiesExecutes Executes
job request job request (workflow)(workflow)
Handles job Handles job requests requests
from usersfrom users
Stores results of computation Stores results of computation after job is executed; enables after job is executed; enables users to retrieve resultsusers to retrieve results
Establishes dynamic data links Establishes dynamic data links between infobetween info--processing & processing & data providing servicesdata providing services
Maintains index of Maintains index of infoinfo--processing processing services registered services registered w/brokerw/broker
EqpmntEqpmnt data: data: nameplate and nameplate and oprtngoprtng, , cndtncndtn, , maintmaint histories. histories. Querying Querying functions.functions.
Data analysis logic: Communicates Data analysis logic: Communicates w/ dataw/ data--providing services in providing services in federated fashion. federated fashion.
99
1010
Layer 1: LongLayer 1: Long--term power system simulationterm power system simulationEMS, Energy management system: in energy control centerEMS, Energy management system: in energy control center
OTS, Operator training simulator: EMS interfaced with OTS, Operator training simulator: EMS interfaced with power system simulator, plus instructor control.power system simulator, plus instructor control.
An $800k project, cost shared between An $800k project, cost shared between –– AREVA ($537k)AREVA ($537k)–– ISU College of ISU College of EngrEngr ($65k)($65k)–– ISU ISU ECpEECpE ($41k)($41k)–– ISU Elect ISU Elect PwrPwr RsrchRsrch CntrCntr ($25k)($25k)–– MidAmerican ($132k)MidAmerican ($132k)
Developed Developed ““Process to ProtectProcess to Protect”” to obtain EMS modelto obtain EMS model
1111
Layer 2: Sensing and Communications•• XfmrXfmr, expensive, winding paper failure ends life., expensive, winding paper failure ends life.•• DGA: Detect gases generated as paper deteriorates. DGA: Detect gases generated as paper deteriorates. •• Purchased DGA monitor at cost ($30k), Purchased DGA monitor at cost ($30k), KelmanKelman donating timedonating time•• MECMEC’’ss sub, sub, DesMoinesDesMoines, monitor 500 MVA, 345/161 kV , monitor 500 MVA, 345/161 kV XfmrXfmr•• Cannon donating cellCannon donating cell--modem, server, communicate realmodem, server, communicate real--time to time to Minneapolis webMinneapolis web--sever, accessed via internetsever, accessed via internet•• Original install date, 10/06, Original install date, 10/06, XfmrXfmr failed day before! Legal issue, failed day before! Legal issue, required contractual agreement. New install date: 7/07, sister urequired contractual agreement. New install date: 7/07, sister unitnit
Photo-Acoustic Spectroscopy for DGA
• Pulsed EM pressure variation, detected using microphones
• Each gas has unique absorption spectrum
• Select f, measure resultant signal, detect concentration of any gas
1212
Layer 2: Sensing and Communications
1313
Layer 3: Data Handling and Integration
ProblemProblem: Decision algorithms need : Decision algorithms need most recent data. Warehoused most recent data. Warehoused data always data always ““oldold”” with respect to with respect to source data.source data.
SolutionSolution: Make data queries : Make data queries directly to source (equipment)directly to source (equipment)
ProblemProblem: Attribute naming : Attribute naming convention varies in source data. convention varies in source data. User must remember names in User must remember names in databases & relation between databases & relation between them.them.
SolutionSolution: Build name mapping : Build name mapping from userfrom user--side to source.side to source.
Implemented these two solutions Implemented these two solutions using a using a federated, queryfederated, query--centric centric data integration systemdata integration system..
Substation 1
Substation 2
Substation 3
Substation 4
Substation n
…
CB
XFMR
CB
XFMR Agent
CB XFMR
CB
XFMR CB
XFMR
Monitored dataMaintenance
histories (corporate database)
Equipment Data
(corporate database)
1414
Data Sources for Layers 3, 4
RealReal--time data available, but only for one transformer, what about ottime data available, but only for one transformer, what about others?hers?
Obtained a $113k contract with MEC to use DGA lab tests to perfoObtained a $113k contract with MEC to use DGA lab tests to perform rm health assessment/ life prediction of entire health assessment/ life prediction of entire xfmrxfmr fleetfleet
WWill use both realill use both real--time & historical data in DDDAStime & historical data in DDDAS
Also use other data sources, weather, other substation equipmentAlso use other data sources, weather, other substation equipment
λ23 λ12 λ34
λjk
deterioration
function g(c(t))
Statistical Processing
Most recent observation
c(T+1)
Maintenance history and test afterwards c(m)
Inspection historical data c(t) t=1,…,T
Failure probability and time to failure
at any time t MinorNew Major Failed
1515
Layer 5: Simulation & Decision: Coupled, Nested: Coupled, NestedOperations T=1-168 hrs
Maintenance T=1-5 yrs
Planning T=5-10 yrs
From→ To ↓ Unit
commit (UC)
Optimal power flow (OPF)
Security Assessmnt(SA)
Shortterm maint
Longterm maint
Investment planning
Unit commit (UC)
Total load
Optimal power flow (OPF)
Units committed
Bus loads, topology
Ope
ratio
ns
Security Assessment (SA)
Units committed
Operating condition
Weather, failure datainst. cndtn data
Shortterm maint
Units committed
Operating condition
Operating (risk) history
Maint effcts, failure data, cdt history, resources
Mai
nten
ance
Longterm maint
Units committed
Operating condition
Operating (risk) history
ST maint schedule, ST eqp deter rate
Cost of capital, failure data cdt history
Plan
ning
Investment planning
Units committed
Operating condition
Operating (risk) history
ST maint schedule, ST eqp. deter rate
LT maint schedule, LT eqp. deter rate
Cost of capital, failure data, cdt history
• Sequential coupling; solution to latter-stage problem depends on former-stage solutions
• Mixed integer
• High dimension
• A good application of Benders decomp
• Problems have not been solved together before; doing so will improve solutions.
1616
Layer 5: Simulation & Decision: Benders: Benders
hFyEx
bAyts
ydxcMin TT
≥+
≥
+
..
:
Problem P
LB=z*
UB=(h-Fy*)Tλ*+dTy*
*FyhEx..
xc min T
−≥ts
Sub-Primal:zmin
Master:
bAy
ydz.. T
≥
≥ts
λTFy*)(h max −Sub-Dual:
cE.. T ≤λts
1. UB=∞
2. Solve Master: y*, LB=z*
3. While UB-LB>ε do:
a. Solve Sub-Dual
b. UB=min{UB,(h-Fy*)Tλ*+dTy*
c. Solve Master with added constraint z≥(h-Fy*)Tλ*+dTy*
d. LB=z*
4. End while
1717
Layer 5: Simulation & Decision: Operational DM: Operational DM
1. Unit commitment: min {Start-up+Shut-down costs} stdemand≤total gen cap≤demand+reserve2. Optimal power flow: min {Cost of generation} stdemand=total gen, line flows≤line capacities3. Security assessment: min {Contingency risk} stdemand=total gen, line flows≤line capacities
Applied to operations:
Applied to maintenance & planning:1. Long-term maintenance: min{InspCst+MntCst+RplcmtCst}2. Investment planning: min {InvstmntCst+PrdCst}
M. Yan, S. Ryan, and J. McCalley, “Transmission Expansion Planning with Transformer Replacement,” Proceedings of the 2007 Industrial Engineering Research Conference, May 19-23, Nashville, TN.
Y. Li, J. McCalley, S. Ryan, “Risk-Based Unit Commitment,” to appear in Proc. of the 2007 IEEE PES General Meeting, June, 2007, Tampa Fl.
1818
Illustration: test systemIllustration: test system
Fig. 4. 6-bus test system Fig. 5. Effect of contingency
0 5 10 15 20 251500
2000
2500
3000
3500
Hours
$
Average10*Average
UC solution is the same in the two cases illustrated in Fig 5.
But UC solution changes if contingency probabilities are 0, an extreme situation which in fact corresponds to the way UC is solved in practice where UC and SA are solved separately.
Better solutions result when different problems are solved together.
1919
ConclusionsConclusionsAreas of work:Areas of work:–– MonitoringMonitoring–– Component life/failure predictionComponent life/failure prediction–– SimulationSimulation
Nested, coupled decision central to power systems, but not Nested, coupled decision central to power systems, but not formally solved in this way due to scaleformally solved in this way due to scale
Capturing influence of one decision problem within another Capturing influence of one decision problem within another improves solution quality of both; decomposition techniques improves solution quality of both; decomposition techniques make it tractablemake it tractable
NSF support being leveraged to bring $400k+$500k more.NSF support being leveraged to bring $400k+$500k more.
–– Software architectureSoftware architecture–– OptimizationOptimization–– Data integrationData integration
2020
Current EffortsCurrent Efforts
• Solve decision problem of installing next monitor
• Couple operational w/ maintenance, planning decision algs
• Install simulator & monitor; implement software design
• Transform monitoring data to failure rates & life prediction
Single bullet: Dynamic-data driven failure indices and coupled decision algorithms improve quality of operational, maintenance, and investment solutions for capital-intensive electric power asset management.