latest computational and mathematical tools for ... · clustering methods: wind regions identified...
TRANSCRIPT
![Page 1: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/1.jpg)
NREL is a national laboratory of the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, operated by the Alliance for Sustainable Energy, LLC.
Latest Computational and Mathematical Tools for Transmission Expansion
IEEE PES T&D Meeting, Chicago IL
Clayton Barrows, PhD
April 15, 2014
![Page 2: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/2.jpg)
2
Grid Decision Horizons
Source: Alexandra von Meier
Planning Operations
“Dynamics”
(Automation)
![Page 3: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/3.jpg)
3
Grid Integration Studies
Scenario Development Penetration of Renewables, Flexibility of Existing Fleet,
Additional Flexibility, Retirements, Firm Capacity
Transmission Expansion (per scenario, iterative to
based on congestion threshold)
System Operation Reserve Requirements,
Dispatch Time Step, Forecast Error Distribution, Market
Horizon
Renewable Resources: Location, Cost,
Production Profile, Capacity Factor
System Model: System, Year, Load
Growth Assumptions, Power Plant Operation
Flexibility: Storage, Demand Side
Resources, Electric Vehicles, etc.
Reliability Stability Flexibility
Production Cost Simulation AC-Optimal Power Flow
Detailed Distribution
System Modeling
Emissions Impact
Costs: Capital, Fuel,
VO&M, Cycling
Barriers Analysis: Regulatory, Economic,
Attitude/Perception
Unit commitment, dispatch, reserve provision, AGC signal, ACE, LMP
(data synthesis by scenario and across simulation horizons (planning, commitment,
dispatch, operation)
Results, Analysis, Knowledge
![Page 4: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/4.jpg)
Planning Models
![Page 5: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/5.jpg)
5
Resource Planning Model (RPM)
• Capacity expansion model for a regional electric system over a utility planning horizon (10-20 year)
• Includes hourly chronological dispatch and detailed system operation representation
• High spatial resolution informs mid- to long-term generator (renewable and non-renewable) siting options
• Nodal (DCOPF) transmission representation with transmission expansion along existing and planned corridors
• Flexible structure to enable easy definition of focus region
Mai, T.; Drury, E.; Eurek, K.; Bodington, N.; Lopez, A.; Perry, A. (2013). Resource Planning Model: An Integrated Resource Planning and Dispatch Tool for Regional Electric Systems. 69 pp.; NREL Report No. TP-6A20-56723.
http://www.nrel.gov/docs/fy13osti/56723.pdf
![Page 6: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/6.jpg)
6
RPM: consistent high resolution data
• 10 km renewable resource grid cells
• Nodal (bus level) power system data
![Page 7: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/7.jpg)
7
Data: Complete Western Interconnection data for all major generation units and transmission lines
1,836 Units
Coal 126
Gas-CC 219
Gas-CT 422
Other Gas 164
Nuclear 8
Hydro 763
Pumped Hydro 15
Geothermal 57
Biomass 62
238 GW
Coal 36
Gas-CC 60
Gas-CT 20
Other Gas 17
Nuclear 10
Hydro 75
Pumped Hydro 4
Geothermal 3.1
Biomass 1.6
21,939 Transmission Lines *In
clu
des
all
un
its
and
lin
es o
nlin
e in
20
10
as
rep
rese
nte
d in
WEC
C T
EPP
C
dat
abas
e u
sed
in W
WSI
S 2
stu
dy.
To
tals
incl
ud
e w
ind
an
d s
ola
r
![Page 8: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/8.jpg)
8
Data: Detail maintained for Rocky Mountain Region, reduced for rest of Western Interconnection
305 Units
Coal 41
Gas-CC 26
Gas-CT 61
Other Gas 8
Nuclear 0
Hydro 164
Pumped Hydro 5
Geothermal 0
Biomass 0
21.8 GW
Coal 6.8
Gas-CC 3.8
Gas-CT 3.4
Other Gas 0.1
Nuclear 0
Hydro 7.1
Pumped Hydro 0.6
Geothermal 0
Biomass 0
1,796 Transmission Lines *In
clu
des
all
un
its
and
lin
es o
nlin
e in
20
10
in R
MP
P a
s re
pre
sen
ted
in W
ECC
TE
PP
C d
atab
ase
use
d in
WW
SIS
2 s
tud
y. T
ota
ls in
clu
de
win
d a
nd
so
lar
31 zones (pseudo-
BAs)
![Page 9: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/9.jpg)
9
Data: Flexible structure enables shift in focus region (e.g. Southwest region for NGS phase 2 study)
187 Units
Coal 22
Gas-CC 42
Gas-CT 53
Other Gas 18
Nuclear 3
Hydro 45
Pumped Hydro 2
Geothermal 0
Biomass 2
35.5 GW
Coal 8.3
Gas-CC 15
Gas-CT 2.4
Other Gas 1.5
Nuclear 4
Hydro 4
Pumped Hydro 0.1
Geothermal 0
Biomass 0.03
1,935 Transmission Lines *In
clu
des
all
un
its
and
lin
es o
nlin
e in
20
10
in S
W r
egio
n a
s re
pre
sen
ted
in
WEC
C T
EPP
C d
atab
ase
use
d in
WW
SIS
2 s
tud
y. T
ota
ls in
clu
de
win
d a
nd
so
lar
![Page 10: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/10.jpg)
10
Data: Flexible structure enables shift in focus region (e.g. CAISO model)
445 Units
Coal 7
Gas-CC 68
Gas-CT 174
Other Gas 66
Nuclear 4
Hydro 58
Pumped Hydro 7
Geothermal 26
Biomass 31
48.6 GW
Coal 0.2
Gas-CC 16.5
Gas-CT 6.5
Other Gas 9
Nuclear 4.5
Hydro 7.4
Pumped Hydro 1.8
Geothermal 2
Biomass 0.6
5,142 Transmission Lines *In
clu
des
all
un
its
and
lin
es o
nlin
e in
20
10
in S
W r
egio
n a
s re
pre
sen
ted
in
WEC
C T
EPP
C d
atab
ase
use
d in
WW
SIS
2 s
tud
y. T
ota
ls in
clu
de
win
d a
nd
so
lar
![Page 11: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/11.jpg)
11
Data clustering: k-means
• Reduce the number of represented data points
o Group similar data points
o Ensure that groups represent similar amounts of data
o … other user defined rules
• K-means clustering
![Page 12: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/12.jpg)
12
Clustering Methods: Solar regions identified based on 8760 profile similarity
• K-means clustering algorithm used instead of arbitrary boundaries
• Higher resolution in focus region (e.g. RMPP)
• Higher resolution in better resource areas (lower latitude)
![Page 13: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/13.jpg)
13
Clustering Methods: Wind regions identified based on annual capacity factors
• WWSIS data and sites used
• Map shows voronoi polygons, but centroid of individual WWSIS sites (within each wind region) used to estimate cost of spur lines
• Consider updating to high resolution 3Tier data and applying clustering methods if time/budget allows
![Page 14: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/14.jpg)
14
Data clustering: Demand
• K-means clustering used to select 3 representative weeks for dispatch modeling
![Page 15: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/15.jpg)
15
High resolution opens opportunities to explore land ownership/management issues
![Page 16: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/16.jpg)
16
Transmission Constraints
• Linearized-DC power flow o Constrained by Kirchhoff's
laws:
𝑓𝑖,𝑗 = 𝑃𝑇𝐷𝐹𝑘,(𝑖,𝑗)𝑠 𝑃𝑘
𝑘
• Transmission constraints: o Inside focus region are defined
by transmission line flow ratings
o Outside focus region are defined by inter-zonal constraints (WECC prescribed corridor limits)
Distribution of flow along linei,j due to a power injection at busk
Power injected at busk
![Page 17: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/17.jpg)
17
Transmission Constraints
• Transmission network reduction: o Outside focus region use
equivalent line representation
o Inside focus region represent each transmission line
o Inside-outside focus region boundary use equivalent line representation
Focus From To Impedance Constraint
Out BA BA Reduced WECC Interface
Border BA Bus Reduced Line Limit
Border Bus BA Reduced Line Limit
In Bus Bus Line Line Limit
![Page 18: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/18.jpg)
18
Transmission Expansion
• Expansion is allowed along existing and “planned” corridors only
𝑓𝑖,𝑗 ≤ 𝑓𝑖,𝑗𝑚𝑎𝑥 + 𝑓𝑖,𝑗
𝑒
o Expansion influences flow ratings (impedances remain static) o Eliminates the requirement to re-calculate power transfer
distribution matrix (just update constraints) o Requires a quantification of uncertainty introduced by these
simplifications. – Is it realistic to assume that re-conductoring/expanding transmission
corridors yields ‘similar’ branch impedances?
• Enables dynamic linear modeling of transmission expansion decisions while retaining DC model: 𝐶𝑖,𝑗
𝑒 ∙ 𝑓𝑖,𝑗𝑒
• Transmission expansion costs derived from existing literature1
1Mills, Andrew D.(2009). The Cost of Transmission for Wind Energy: A Review of Transmission Planning Studies. Lawrence Berkeley National Laboratory: Lawrence Berkeley National Laboratory. Retrieved from: http://escholarship.org/uc/item/2z12z7wm
![Page 19: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/19.jpg)
19
Location Dependent Results
• Operational value
o Energy markets
• Capacity value
o Reliability standards
o Renewable portfolios
![Page 20: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/20.jpg)
Operations Simulations
![Page 21: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/21.jpg)
21
Unit commitment and economic dispatch
optimization horizon: 48 hours
![Page 22: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/22.jpg)
22
Unit commitment and economic dispatch
rolling forward in 24 hour increments
![Page 23: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/23.jpg)
23
Unit commitment and economic dispatch
The state of the system at time t=0 is dependent on:
1. Generator commitment status: on/off 2. If “on”: hours of continuous operation;
current ramp rate 3. If “off”: hours since last operation
(minimum shut down duration)
![Page 24: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/24.jpg)
24
Unit commitment and economic dispatch
Each optimization problem takes between 2 minutes and 7 hours to solve. Annual solutions can range from hours to weeks.
PLEXOS
![Page 25: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/25.jpg)
25
Changes to generator configuration…
• Annual solutions are important because load and renewable generation have seasonal variations.
• Increasing renewable penetration increases the number and magnitude of short duration ramps. • Investment decisions are based on both investment costs and long–term operational costs &
savings. Production cost models are the choice tool for evaluating the majority of operational costs & savings, however they do not estimate make investment decisions.
![Page 26: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/26.jpg)
26
Eastern Renewable Generation Integration Study (ERGIS)
• 6,500 Generators o ~850 GW Capacity
o 3,300 TWh
• 62,000 Busses
• 57,000 Transmission Lines
• 23,000 Transformers
• ~850GW Capacity
26
![Page 27: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/27.jpg)
27
Inter-Area Transmission
![Page 28: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/28.jpg)
28
UC on NREL’s HPC
Peregrine Characteristics: • 11520 Intel Xeon E5-2670 "SandyBridge" cores • 14400 next-generation Intel Xeon "Ivy Bridge"
core • 576 Intel Phi Intel Many Integrated Core (MIC)
core co-processors with 60+ cores each • 32 GB DDR3 1600Mhz memory per node • Peregrine will deliver a peak performance of 1
petaFLOPS
NR
EL P
IX 2
45
80
Domain
Decomposition
PCM
Solution
Analysis
Partitioning
PCM PCM PCM
Solution Solution
Reassembly
PLEXOS PLEXOS PLEXOS
![Page 29: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/29.jpg)
29
Change in UC/ED with no overlap
Started UC/ED on January 31 at 12:00 am
Started UC/ED on January 1 at 12:00 am
![Page 30: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/30.jpg)
30
Idea: Parallelize in the time domain
Hypothesis: a decision at time t is not dependent on the state of the system at previous time intervals, given a delay of n time periods.
Plotted here: Autocorrelation of the generator unit commitment decision variable for a group of generators.
The duration of the lag necessary for the autocorrelation of the Unit Commitment to reach a local minimum is called the Unit Commitment Decision Persistence, or just Persistence
![Page 31: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/31.jpg)
31
Effect of Overlap on Dispatch
Normalized root mean square difference (NRMSD) in generation dispatch, by type of generator, relative to the annual solution. This calculation is made each day and plotted relative to the number of overlap days (number of days since the start of the optimization).
![Page 32: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/32.jpg)
32
ST Solution Time
9-day simulations: weekly with 2-days of overlap.
![Page 33: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/33.jpg)
33
Speedup in hourly UC simulation (RMPP) Annual solution takes 131.7 minutes. With 52 partitions (with increasing overlap days):
52
par
alle
l 7-d
ay s
imu
lati
on
s
52
par
alle
l 8-d
ay s
imu
lati
on
s
52
par
alle
l 9-d
ay s
imu
lati
on
s
(4.3 minutes)
(4.4 minutes)
(5.1 minutes)
(5.3 minutes)
(6.0 minutes)
(6.6 minutes)
![Page 34: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/34.jpg)
34
ERGIS Simulation Time Comparison
Windows 12 Partitions: 155 hr (6.4 days)
HPC 52 Partitions: 64 hr (2.7 days)
Windows 1 Partition: 1,610 hr (67 days)
![Page 35: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/35.jpg)
35
Conclusions
• Many operations and planning problems remain computationally intractable
• Simplifications can be made through constraint relaxation or transmission network equivalencing o Current methods may not be robust to network
augmentation
• Ongoing work: o Heuristic techniques to identify relevant constraints
o Further parallelism through spatial domain decomposition
![Page 36: Latest Computational and Mathematical Tools for ... · Clustering Methods: Wind regions identified based on annual capacity factors •WWSIS data and sites used •Map shows voronoi](https://reader033.vdocuments.mx/reader033/viewer/2022053123/60ac8ec13b038971226fccb5/html5/thumbnails/36.jpg)
36
Contact:
Clayton Barrows
National Renewable Energy Laboratory
Energy Forecasting and Modeling Group
Strategic Energy Analysis Center
303-275-3921
Questions?