Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Generating Realistic Information for the Development of
Distribution And Transmission Algorithms
GRID DATA Program Introduction
Tim HeidelProgram Director
Advanced Research Projects Agency – Energy (ARPA-E)
U.S. Department of Energy
GRID DATA Kickoff Meeting
Denver, CO, March 30-31, 2016
Emerging Grid Challenges
1
– Increasing wind and solar
generation
– Decentralization of generation
– Aging infrastructure
– Changing demand profiles
– Increasing natural gas generation
– Cybersecurity threats
‣ All of these challenges require
new tools for faster, better,
more robust grid optimization.
Responsive Demands
- Scheduling large loads (eg. industrial loads)
- Mobilize large numbers of small assets
Power Flow Controllers
- AC Power Flow Controllers
- High Voltage DC Systems
Energy Storage Optimization
- Scheduling energy flows
- Coordination of diverse storage assets
Transmission Topology Optimization
- Optimal line switching
- Corrective switching actions
‣ Advances in power electronics, computational technologies, and
mathematics offer new opportunities for optimizing grid operations.
New Opportunities for Grid Optimization
2
CD-PAR CAD Image
115kV, 1500A Prototype (2-5 Ω)
Continuously Variable Series Reactor
50uH (<150 lbs) Prototype
Distributed Series Reactor
3
ARPA-E GENI Optimization Projects (2012-2015)
4
AC-Optimal
Power Flow
Stochastic
Optimization
Distributed
Optimization
Optimal
Forecasting &
Dispatch of
Demand
Transmission
Switching
Energy
Storage
Optimization
Fast
voltage/transi
ent stability
calculations
Grid
Optimization
Toolkit
Advanced Computing Cost Reductions & Performance Gains
New Optimization Methodologies & Advanced Solvers
PMU-Based
State
Estimation
Fully leveraging GENI
successes will require new
Optimal Power Flow (OPF) tools…
5
But GRID DATA is not focused on
OPF algorithm development…
Primary Technical Targets (from GENI Funding Opportunity Announcement)
SCALABILITY: Capable of managing large dynamical systems (>10,000 nodes)
VALIDATION: Real-world datasets supplied by transmission operators or utilities.
FEASIBILITY: Consideration of sensing, communications, computational, and
actuation (ramp and dispatch) challenges for implementation in “real-time” markets.
FAILSAFE: Designs where a safe, “dumb” operation occurs in the event of local or
wide- area failure or attack.
GENI Program Targets
6
Exemplary GENI Modeling Results
7
Boston University & PJM
(Topology Control)
• PJM day-ahead + real-time simulations based on
historical data (2010, 2013, 2014).
• Generation economics.
• Must-run, maintenance and outage schedules.
• Load profiles and forecasts.
• Reserve requirements.
• Operational power flows (inc. historical topology).
• Interchange with neighboring regions.
• Transmission constraints and contingencies.
• > 13,000 nodes (up-to 100k for breaker node), >
18,000 branches, > 6000 single and multi-element
contingencies.
• Benchmarked both DA and RT market results against
historical data (nodal prices (major trading hubs),
dispatched generation mix, congestion costs and
congestion patterns).
Caltech & Southern
California Edison
(Distributed AC-OPF)
• SCE distribution systems
• 6 feeders (4KV, 12KV)
• ~15,000 buses
• <10% error compared with
substation measurements
Challenges with Requiring Real Datasets
‣ Realistic, large-scale datasets are extremely valuable but also difficult,
time consuming and expensive to collect, prepare, and use.
– Every team must negotiate unique data agreement.
– Base cases from ISO/utilities usually do not converge (substantial
cleaning always required).
‣ Data typically cannot be published in any form.
– Very difficult to independently verify/replicate results.
– Results may reflect quality of data more than quality of algorithms.
‣ ISOs/utilities have limited bandwidth to devote to R&D.
– Very few teams can put together credible project plans up front.
– High barrier to entry for those not already in power systems field.
8
Public Benchmark Power System Models
‣ Existing datasets are not adequate
‣ There are too few of them
‣ They are too small
‣ They are not representative of real systems
‣ They are incomplete
‣ They are too easy
Public OPF test systems are drawn from:
• IEEE Power Flow, Dynamic and Reliability,
MATPOWER, Edinburgh, EIRGrid, Other Publication
Test Cases
There are approximately 35
widely available public datasets. IEEE 30 bus.
9
Public OPF test systems are drawn from:
• IEEE Power Flow, Dynamic and Reliability,
MATPOWER, Edinburgh, EIRGrid, Other Publication
Test Cases
There are too few existing public datasets
• Compare against another data intensive field,
computer vision:
• Caltech 101: 9146 images
• Caltech 256: 30,607 images
• LabelMe: 106,739 images
• OPF is solved each year: 1 hour snapshots for a
year = 8,760 datasets for a single system.
IEEE 30 bus.
Images from Caltech 256.
There are approximately 35
widely available public datasets.
10
Existing public datasets are too small
• Real transmission networks are 5,000-50,000 buses.
• Almost all test cases comprise less than 4,000 buses.
Eastern Interconnection Transmission
Network (100kV+ only) Distribution of public test case sizes
11
Not representative of real systems (Examples)
‣ Extremely large (typically unobserved) voltage drops
‣ Low base voltages and an overabundance of voltage control capacity
‣ Lines with non-physical negative resistances (due to undocumented
network reductions).
‣ Lines with non-zero MW thermal emergency ratings, zero MW normal
ratings.
‣ All generators of each type have equivalent characteristics (and cost
curves).
‣ Identical subnetworks are repeated multiple times.
‣ Lists of contingencies, emergency (short term) equipment ratings,
protection system details, generator ramp rates and real and reactive
capability curves, transformer tap settings, capacitor bank locations and
settings, phase shifting transformer characteristics, energy storage
capacity, line switching capabilities, and flexible demand are more often
than not omitted.
Existing public datasets are incomplete
13
• To have any hope of replicating
a real-world OPF problem,
dataset must (at a minimum)
include:
• Generator capabilities
• Generator costs
• Thermal line limits
• Most existing test sets lack
information on these key features.
Right: Datasets listed.
Spaces with “-” indicate
missing information from
the dataset.
C. Coffrin et al.NESTA: The Nicta Energy
System Test Case Archive, arXiv preprint
arXiv:1411.0359v1 (2014) 13
Existing Datasets Are Too Easy (?)
‣ In theory, to find a global solution
can take a time exponential in
the size of the network.
‣ In practice, existing solvers
and/or heuristics find solutions to
the existing test sets that are
extremely close to globally
optimal solutions very quickly.
– Line thermal and generator limits
are set to large non-binding values.
– Generators assigned quadratic cost
curves, often with the same
coefficients Low optimality gaps indicate that heuristics
are extremely close to the global optimum
(red). Surprisingly, the potential gains from
line switching are provably low (blue).
14C. Coffrin et al. NESTA: The Nicta Energy System Test Case Archive, arXiv
preprint arXiv:1411.0359v1 (2014)
Network Connectivity
Line Thermal Limits
SS Generator Characteristics
Generator Cost Curves
Time Series Load Data (by bus)
Contingency Lists
Bus shunt/transformer tap settings
Normal/Emergency Ratings
Dynamic Generator Characteristics
Maintenance Outages
Automated Local Controls
Protection Settings/Coordination
Power Market Design Details
Operator actions (intuition)
GRID DATA Program Objective
Accelerating the development, evaluation, and adoption of new grid
optimization algorithms will require more realistic, detailed public datasets.
GRID DATA:
Increasing
Complexity &
Completeness
Current
datasets
15
GRID DATA:
Increasing
Realism
“Realistic
but not
Real”
Two Pathways to New Datasets
Real Data
- Start with real data, then anonymize,
perturb topologies and change
sensitive infrastructure asset data as
necessary.
- Risks:
- Requires extremely close collaboration
with ISOs such that infrastructure is not
reconstructable and can be publically
released.
- Datasets may no longer well represent
real data.
- Real data is often messy, incomplete.
Open-access, large, realistic,
validated datasets
Synthetic Data
- Generate via expert input,
geographic/road data and data mining.
- Generate new random graph methods
for transmission networks.
- Devise statistical metrics (moments of
capacity distributions, degree
distributions of networks); validate
against real data.
- Risks:
- Validation metrics may be incomplete or
misleading. (Leading to lack of realism.)
16
New Model Repositories Needed
17
Existing mechanisms for
sharing and collaboratively
developing, reviewing
models are limited.
New Model Repositories Needed
18
‣ Enhance research repeatability (and transparency) by enabling the
collaborative maintenance and version control of models.
‣ Researchers need to be able to easily contribute and share new models
with the community.
‣ Open source software development community has enabled highly
productive, widely distributed, technical collaboration involving
thousands of individuals.
Project Categories
GoalsDuration 2016-2018
Projects 7
Total
Investment$11 Million
Program
DirectorDr. Tim Heidel
GRID DATA ProgramGenerating Realistic Information for the Development
of Distribution And Transmission Algorithms
Development of large-scale, realistic,
validated, and open-access electric power
system network models with the detail
required for successful development and
testing of new power system optimization
and control algorithms.
• Transmission, Distribution, and Hybrid Power System Models & Scenarios
• Models derived from anonymized/obfuscated data provided by industry partners
• Synthetic models (matching statistical characteristics of real world systems)
• Power System Model Repositories
• Enabling the collaborative design, use, annotation, and archiving of R&D models
19
Power System Network Model Requirements
‣ Teams may choose to address any specific OPF application(s).
‣ Any method(s) may be used to create test systems (using real-world data
or purely synthetic approaches).
‣ Teams may choose to address (i) transmission/bulk power systems, (ii)
distribution systems, or (iii) hybrid transmission and distribution systems.
‣ Required and optional model details were specified in the FOA.
‣ Detailed plan for validation with technical success/fail criteria required.
‣ Models must be publicly releasable and must not contain CEII data.
20
Transmission
At least one small network model having between 50 and 250 electrical
buses required and at least one large network model having > 5,000
buses. (Larger test systems may not consist of repeated duplicates of
smaller systems.)
Distribution
At least one model with at least 3 independent feeders originating at
one or more substations, corresponding to a minimum of at least 5,000
individual customers.
Scenario Creation Requirements
21
‣ Scenario sets must be designed with temporal resolutions and time-
coupling suitable for solving one or more specific OPF problems.
‣ Any method(s) may be used to create power system scenarios (using
real-world data or purely synthetic approaches).
‣ Teams must generate at least a full year of time-coupled physically
feasible scenarios with at least hourly granularity. (Teams are strongly
encouraged to use the shortest feasible time step between scenarios (5
minutes, 15 minutes, etc.)).
‣ Scenarios must represent a range of difficulty to OPF optimization
algorithms. Teams are also encouraged to develop infeasible scenarios
(to test the ability for OPF algorithms to identify infeasibility quickly).
‣ Required and optional scenario details were described in the FOA.
‣ Teams must have a detailed plan for validation with technical success/fail
criteria to ensure scenarios are sufficiently representative of a range of
real-world power system operating conditions.
Repository Creation Requirements
‣ The repository must be completely open (including international access),
giving researchers the ability to upload modified versions of existing
models and designate relationships between different models (i.e.
version control) as well as provide annotation and/or comments on
specific models (similar to, for example, GitHub).
‣ The repository should be able to accommodate different kinds of power
system models (not just ones suitable for OPF control and optimization).
‣ The repository should have the ability to scale the repository to archive
an arbitrary number of power system models.
‣ Teams have proposed a self-funding mechanism with potential to extend
well beyond ARPA-E’s development funding.
‣ Teams are required to establish a set of standards for models and a clear
self-governance model for the repositories.
‣ The teams must design a plan for active curation of power system
models in the repositories.
22
GRID DATA Project Portfolio
23
Power System Models & Scenarios Model Repositories
T Transmission Models
D Distribution Models
H Hybrid Models
T
PI: Prof. C. DeMarco
T
PI: Prof. T. Overbye
H
PI: Dr. H. HuangPIs: Dr. B. Hodge
& Dr. B. Palmintier
D
T
PI: Prof. P. Van Hentenryck
PI: Dr. M. Rice
PI: Dr. A. Vojdani
23
GRID DATA Program Participants
24
Lead Organizations
Subs/Team Members
CIT
GridBright
Avista
ASU
NRECA
PJM
MIT
PNNL
VCU
CAISO
University of
Wisconsin
Columbia Univ
University
of Michigan
UISOL
UIUC
LANL
Cornell
ANL
GE/Alstom
ComEd
GAMS
NREL
Additional ARPA-E Performer Presentations
25
OPEN 2012
OPEN 2012
OPEN 2012
Cyber-Physical Modeling and
Analysis for a Smart and
Resilient Grid
PI: Prof. Pete Sauer
Non-Wire Methods for
Transmission Congestion
Management through Predictive
Simulation and Optimization
PI: Dr. Henry Huang
Micro-Synchrophasors for
Distribution Systems
PI: Dr. Alexandra von Meier
Additional ARPA-E Performer Presentations
26
IDEAS
OPEN 2015
OPEN 2015
Coordinated Operation of Electric And Natural Gas Supply Networks: Optimization Processes And Market DesignPI: Dr. Alex Rudkevich
High Performance Power-grid Optimization (HIPPO) for Flexible and Reliable Resource Commitment Against Uncertainties
PI: Dr. Feng Pan
Global-Optimal Power Flow (G-OPF)
PI: Prof. Hsiao-Dong Chiang
Bigwood Systems Inc.
Kickoff Meeting Objectives
27
Research
Teams
Industry
Experts
Gov’t
PMs
ARPA-E
Knowledge
Cross-disciplinary
learning about issues
and opportunities
Learning and
industry insights for
researchers & gov’t
Summary of program
goals
Relationships
Potential collaborations
between research
teams
Industry engagement to
improve and gain
access to research
outcomes
Future development
opportunities within
industry and gov’tCommunity
Datasets
making an
impact in the
worldideas
x
xx
incomplete team
uncertain value
poor implementation
Ideas alone are often not enough
Low yield
28
Datasets
making an
impact in the
world
ideas+ value (Techno-economic analysis)
+ team (Stakeholder engagement)
+ implementation (Skills and Resources)
ARPA-E Tech-to-Market tries to improve yield
29
GRID DATA Kickoff Meeting Objectives
30
‣ Discuss GRID DATA objectives (especially first year goals)
‣ Provide critical feedback on approaches and applications.
‣ Explore partnership opportunities and potential synergies.
‣ Brainstorm strategies for maximizing GRID DATA impact
‣ Discuss ARPA-E OPF competition vision
Agenda: Wednesday Morning
31
Start Time
Institution/Presenter Project Title
DAY 1
8:00 Eric Rohlfing (ARPA-E) Welcome and Introductions
8:15 Tim Heidel (ARPA-E) GRID DATA Program Introduction
9:00 Richard O’Neill (FERC) Generating Good Test Problems
9:20 Yonghong Chen (MISO) Bridging a Gap: The Role of and Challenges for GRID DATA
9:40 Networking Break
GRID DATA Model Development
10:10 Wisconsin (GRID DATA)EPIGRIDS: Electric Power Infrastructure & Grid Representation in Interoperable Data Set
10:30 Michigan (GRID DATA)High Fidelity, Year Long Power Network Data Sets for Replicable Power System Research
10:50 UIUC (GRID DATA) Synthetic Data for Power Grid R&D
11:10 PNNL (GRID DATA)Sustainable Data Evolution Technology (SDET) for Power Grid Optimization
11:30 NREL (GRID DATA)SMARtDaTa: Standardized multi-scale Models of Anonymized Realistic Distribution and Transmission data
11:50 Discussion Moderator: Marija Ilic
12:20 LUNCH
Agenda: Wednesday Afternoon
32
Start Time
Institution/Presenter Project Title
DAY 1
12:20 LUNCH
13:20GRID DATA BREAKOUT SESSION #1:
Model Validation
14:40 Networking Break
15:10 BREAKOUT SESSION #1 Reports
OPF Competition
15:40 Tim Heidel (ARPA-E) OPF Competition Introduction and Overview
16:10 PNNL (OPF Competition) ARPA-E Power Grid Optimization Competition Design
16:55 Discussion
17:15POSTER SESSION
Agenda: Thursday Morning
33
Start Time
Institution/Presenter Project Title
DAY 2
8:30 Tim Heidel (ARPA-E) Welcome and Recap
8:40 Patrick Panciatici (RTE)Realistic data for challenging problems; an internal TSO R&D perspective
9:00 Tao Hong (UNC Charlotte) Lessons learned from organizing energy forecasting competitions
GRID DATA Repository Development
9:20 PNNL (GRID DATA)Data Repository for Power system Open models With Evolving Resources (DR POWER)
9:40 GridBright (GRID DATA)A Standards-Based Intelligent Repositorty for Collaborative Grid Model Management
10:00 Discussion Moderator: Carleton Coffrin
10:20 Networking Break
10:50GRID DATA BREAKOUT SESSION #2:
Data Formats and Accelerating Adoption
12:05 LUNCH
13:00 BREAKOUT SESSION #2 Reports
Agenda: Thursday Afternoon
34
Start Time
Institution/Presenter Project Title
DAY 1
OPEN FOA 2012 Projects
13:30 CIEE (OPEN 2012) Micro-Synchrophasors for Distribution Systems
13:50 UIUC (OPEN 2012)Cyber-Physical Modeling and Analysis for a Smart and Resilient Grid
14:10 PNNL (OPEN 2012)Non-Wire Methods for Transmission Congestion Management through Predictive Simulation and Optimization
14:30 Discussion Moderator: Terry Oliver
14:50 Networking Break
OPEN FOA 2015 & IDEAS Projects
15:20Newton Energy Group
(OPEN 2015)Coordinated Operation of Electric And Natural Gas Supply Networks: Optimization Processes And Market Design
15:40 PNNL (OPEN 2015)High Performance Power-grid Optimization (HIPPO) for Flexible and Reliable Resource Commitment Against Uncertainties
16:00 Bigwood Systems (IDEAS) Global-Optimal Power Flow (G-OPF)
16:20 Discussion Moderator: Josh Gould
16:45 Final Discussion Program Director Wrap-up
35
www.arpa-e.energy.gov
Tim Heidel
Program Director
Advanced Research Projects Agency – Energy (ARPA-E)
U.S. Department of Energy