NREL is a national laboratory of the U.S. Department of Energy
Office of Energy Efficiency & Renewable Energy
Operated by the Alliance for Sustainable Energy, LLC
This report is available at no cost from the National Renewable Energy
Laboratory (NREL) at www.nrel.gov/publications.
Contract No. DE-AC36-08GO28308
Variable Renewable Energy in
Long
-Term Planning Models:
A
Multi-Model Perspective
Wesley
Cole, Bethany Frew, Trieu Mai,
and
Yinong Sun
National Renewable Energy Laboratory
John
Bistline, Geoffrey Blanford,
and
David Young
Electric Power Research Institute
Cara Marcy and Chris Namovicz
U.S.
Energy Information Administration
Risa Edelman, Bill Meroney, Ryan Sims,
and
Jeb Stenhouse
U.S.
Environmental Protection Agency
Paul Donohoo
-Vallett
U.S. Department of Energy
Technical Report
NREL/TP-6A20-70528
November 2017
NREL is a national laboratory of the U.S. Department of Energy
Office of Energy Efficiency & Renewable Energy
Operated by the Alliance for Sustainable Energy, LLC
This report is available at no cost from the National Renewable Energy
Laboratory (NREL) at www.nrel.gov/publications.
Contract No. DE-AC36-08GO28308
National Renewable Energy Laboratory
15013 Denver
West Parkway
Golden, CO 80401
303-275-3000 • www.nrel.gov
Variable Renewable Energy in
Long-term Planning Models:
A Multi-model Perspective
Wesley Cole, Bethany Frew, Trieu Mai,
and Yinong Sun
National Renewable Energy Laboratory
John Bistline, Geoffrey Blanford,
and David Young
Electric Power Research Institute
Cara Marcy and Chris Namovicz
U.S. Energy Information Administration
Risa Edelman, Bill Meroney, Ryan Sims,
and Jeb Stenhouse
U.S. Environmental Protection Agency
Paul Donohoo-Vallett
U.S. Department of Energy
Prepared under Task No. OOSP.10291.17.02.25
Technical Report
NREL/TP-6A20-70528
November 2017
NOTICE
This report was prepared as an account of work sponsored by an agency of the United States government.
Neither the United States government nor any agency thereof, nor any of their employees, makes any warranty,
express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of
any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately
owned rights. Reference herein to any specific commercial product, process, or service by trade name,
trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation,
or favoring by the United States government or any agency thereof. The views and opinions of authors
expressed herein do not necessarily state or reflect those of the United States government or any agency thereof.
This report is available at no cost from the National Renewable Energy
Laboratory (NREL) at www.nrel.gov/publications.
Available electronically at SciTech Connect http:/www.osti.gov/scitech
Available for a processing fee to U.S. Department of Energy
and its contractors, in paper, from:
U.S. Department of Energy
Office of Scientific and Technical Information
P.O. Box 62
Oak Ridge, TN 37831-0062
OSTI http://www.osti.gov
Phone: 865.576.8401
Fax: 865.576.5728
Email: [email protected]
Available for sale to the public, in paper, from:
U.S. Department of Commerce
National Technical Information Service
5301 Shawnee Road
Alexandria, VA 22312
NTIS http://www.ntis.gov
Phone: 800.553.6847 or 703.605.6000
Fax: 703.605.6900
Email: [email protected]
Cover Photos by Dennis Schroeder: (left to right) NREL 26173, NREL 18302, NREL 19758, NREL 29642, NREL 19795.
NREL prints on paper that contains recycled content.
iii
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Acknowledgments
We gratefully acknowledge the many people whose efforts contributed to this report. The
IPM, NEMS, ReEDS, and US-REGEN modeling teams all contributed to the overall model
development and the discussion of ideas presented in this work. Steve Capanna (DOE) and
David Hunter (EPRI) were instrumental in designing this project and in organizing the
workshops. We are grateful to Evelyn Wright (Sustainable Energy Economics) and Jeff Logan
(NREL) for providing feedback on this work. The work for this report was funded by the EERE
Office of Strategic Programs, Solar Energy Technology Office, and Wind Energy Technology
Office under contract number DE-AC36-08GO28308, and by the Environmental Protection
Agency. The views and opinions expressed in this paper are those of the authors alone and do
not necessarily reflect those of the U.S. Government, the Environmental Protection Agency, the
Energy Information Administration, the National Renewable Energy Laboratory, or the Electric
Power Research Institute, and no official endorsement should be inferred.
iv
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
List of Acronyms
AEO Annual Energy Outlook
CEM capacity expansion model
CSAPR Cross-State Air Pollution Rule
CSP concentrating solar power
CV capacity value
EIA U.S. Energy Information Administration
EMM Electricity Market Module
EPA Environmental Protection Agency
EPRI Electric Power Research Institute
ERCOT Electric Reliability Council of Texas
GW gigawatt
HIVG high variable generation
IPM Integrated Planning Model
kW kilowatt
LCOE levelized cost of electricity
LDC load duration curve
LP linear program
MATS Mercury and Air Toxics Standards
MW megawatt
NEMS National Energy Modeling System
NERC North American Electric Reliability Corporation
NLDC net load duration curve
NPV net present value
O&M operation and maintenance
PV photovoltaic
QCP quadratically constrained program
R&D research and development
RE renewable energy
ReEDS Regional Energy Deployment System
RPS renewable portfolio standard
TWh terawatt-hours
US-REGEN United States Regional Economy, Greenhouse Gas, and Energy
VRE variable renewable energy
WACC Weighted Average Capital Cost
v
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Executive Summary
Long-term capacity expansion models of the U.S. electricity sector have long been used to
inform electric sector stakeholders and decision makers. With the recent surge in variable
renewable energy (VRE) generators—primarily wind and solar photovoltaics—the need
to appropriately represent VRE generators in these long-term models has increased. VRE
generators are especially difficult to represent for a variety of reasons, including their variability,
uncertainty, and spatial diversity. To assess current best practices, share methods and data, and
identify future research needs for VRE representation in capacity expansion models, four
capacity expansion modeling teams from the Electric Power Research Institute, the U.S. Energy
Information Administration, the U.S. Environmental Protection Agency, and the National
Renewable Energy Laboratory conducted two workshops of VRE modeling for national-scale
capacity expansion models. The workshops covered a wide range of VRE topics, including
transmission and VRE resource data, VRE capacity value, dispatch and operational modeling,
distributed generation, and temporal and spatial resolution. The objectives of the workshops were
both to better understand these topics and to improve the representation of VRE across the suite
of models. Given these goals, each team incorporated model updates and performed additional
analyses between the first and second workshops. This report summarizes the analyses and
model “experiments” that were conducted as part of these workshops as well as the various
methods for treating VRE among the four modeling teams. The report also reviews the findings
and learnings from the two workshops. We emphasize the areas where there is still need for
additional research and development on analysis tools to incorporate VRE into long-term
planning and decision-making.
Note: This research is intended to inform the energy modeling community on the modeling of
variable renewable resources, and is not intended to advocate for or against any particular
energy technologies, resources, or policies. Scenarios evaluated as part of this work were
selected to exercise specific model capabilities, and do not reflect policy preferences or market
expectations of the participating organizations or modelers.
vi
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table of Contents
1 Introduction ........................................................................................................................................... 1
2 Model Summaries ................................................................................................................................. 5
3 Results and Discussion ....................................................................................................................... 9
3.1 Spatial and Temporal Resolution .................................................................................................. 9
3.1.1 Examples of Temporal Resolution Methods .................................................................. 12
3.1.2 Examples of Spatial Resolution Methods ...................................................................... 15
3.1.3 Recommendations for Future Modeling R&D ............................................................... 18
3.2 Resource Adequacy ..................................................................................................................... 19
3.2.1 VRE Contributions to Resource Adequacy .................................................................... 20
3.2.2 Determination of the Resource Adequacy Level ........................................................... 22
3.2.3 Other Resource Adequacy Challenges ........................................................................... 23
3.2.4 Recommendations for Future Modeling R&D ............................................................... 24
3.3 Economics of Energy Production ................................................................................................ 24
3.3.1 Recommendations for Future Modeling R&D ............................................................... 28
3.4 Other Considerations ................................................................................................................... 29
4 Summary and Conclusions ............................................................................................................... 30
References ................................................................................................................................................. 31
Appendix. Summary of Model Enhancements ....................................................................................... 35
IPM ..................................................................................................................................................... 35
NEMS ................................................................................................................................................... 35
ReEDS .................................................................................................................................................. 35
US-REGEN .......................................................................................................................................... 35
vii
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
List of Figures
Figure 1. Native spatial resolution of the four models: 64 IPM regions (top left), 22 NEMS
regions (top right), 134 ReEDS balancing area regions (bottom left), and 48 US-REGEN states
(bottom right) .............................................................................................................................. 10
Figure 2. NEMS deployment results with (“w/ Curt”) and without (“No Curt”) 864 Curtailment
method using 9 versus 15 time slices .......................................................................................... 13
Figure 3. ReEDS LDC-based approach to calculating CV .................................................................. 14
Figure 4. Incremental PV CV using the former and new ReEDS CV method in the Austin, Texas,
region (left) and Southern California region (right) .................................................................... 15
Figure 5. Annual production cost savings from 22 aggregated transmission regions (versus 64 native
regions) for sample year 2040 ..................................................................................................... 16
Figure 6. Incremental compliance cost of a 50% wind and solar RPS (relative to a reference scenario
without these mandates) with 15 versus 48 US-REGEN regions (Bistline et al. 2017) .............. 17
Figure 7. Reserve margins for each NERC assessment area (NERC 2017) ......................................... 20
Figure 8. Capacity value for wind (left) and PV (right) as a function of PV and wind penetration ..... 22
Figure 9. Coal capacity (left) and nuclear capacity (right) in the model’s reference case scenarios .... 23
Figure 10. NEMS generation mix in the reference case when curtailment is included (left) and the
difference in generation when curtailment is not included (right) .............................................. 26
Figure 11. Impact of financing assumptions on the LCOE of a wind plant ......................................... 27
List of Tables
Table 1. Summary of VRE Attributes Relevant to Capacity Expansion Models ................................... 2
Table 2. Model and Computational Details ............................................................................................ 6
Table 3. Power Sector Constraints/Implementation ............................................................................... 7
Table 4. VRE-Specific Characteristics ................................................................................................... 8
Table 5. Summary of Temporal Resolution and Methods for the Four Models ................................... 11
Table 6. ReEDS’ Total System Cost with 134 (REF), 48 (STATE), and 13 (NERC) Regions
for VRE Representation (Krishnan and Cole 2016) .................................................................... 18
Table 7. Summary of How Conventional and VRE Capacity Contribute to Resource Adequacy for the
Four Models ................................................................................................................................ 21
Table 8. Financing Assumptions Used in the Four Models ................................................................. 27
Table 9. Summary of Transmission Connection Costs for VRE Technologies ................................... 28
1
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
1 Introduction
Capacity expansion models (CEMs) have long been used as tools by electric utility planners,
policymakers, and stakeholders to inform decisions related to the evolution of the power sector.
These models typically evaluate the least-cost portfolio of electricity generators, transmission,
and storage needed to reliably serve load over many years or decades. By evaluating different
possible future scenarios, CEMs can help identify key drivers of change and assess the
implications of different generation portfolios.
CEMs vary widely in their geographic scopes, which inform tradeoffs with temporal and spatial
resolution modeling choices. For example, CEMs are used to evaluate systems ranging from
local systems with tens of megawatts (MW) to national systems with greater than 1,000
gigawatts (GW), such as the four models included in this report. International versions have
been used in the integrated assessment model community to evaluate optimal pathways for
decarbonizing the global electricity system (e.g., Pietzcker et al. 2014; Pietzcker et al. 2017;
IPCC 2015; Edelenbosch et al. 2017). In this report, we take a relatively narrow focus and
consider only national-scale models of the U.S. power sector,
1
as this scope is commonly used
to inform national-level decisions for various government and private sector entities. These
national-scale models are not as large as integrated assessment models, which generally require
more simplified temporal and spatial treatment to accommodate greater model complexity when
they include all energy sectors and carriers, all world regions, and the full 21
st
century. On the
other hand, there are regional models with higher resolution but a much narrower scope (e.g.,
WECC 2013; ABB 2016; Mai et al. 2015; Nelson et al. 2012). These models are typically used
for different applications
2
than the four models in this project and thus are not fully comparable.
We do not assess these other classes of models in this work.
Variable renewable energy (VRE) is one of the major sources of complexity in national-scale
CEMs because of its physical attributes, which differ from conventional generator technologies,
and the need to more accurately represent those attributes. VRE technologies rely on a renewable
fuel source and are typically non-dispatchable; in the case of this report, we limit VRE
technologies to those that depend on solar and wind generation.
3
With the rapid increase in VRE
deployment over the past decade, the need to accurately represent VRE in CEMs has grown in
importance. However, VRE technologies have many differences from conventional technologies,
such as natural gas, coal, and nuclear units, which make them more challenging to incorporate
into CEMs. Table 2 summarizes many of these VRE attributes, which we compiled from various
sources, including Milligan et al. (2016) and Kroposki et al. (2017). These attributes, to varying
degrees, have economic implications for VRE, other electric sector investments, and system
operations (Ueckerdt et al. 2013; Blanford et al. 2016). For example, VRE plant output is
weather-driven and varies considerably from one location to another, while conventional
1
Some national-scale U.S. models also represent portions of the Canadian power system because of the synchronous
interconnection ties across the two power systems. Mexican power plants that are part of the Western Electricity
Coordinating Council might also be included.
2
For example, higher-resolution regional models are often used to evaluate individual projects, such as a new plant
addition, a new transmission line, or the retirement of a plant.
3
Although other renewable technologies have variable output, such as conventional hydroelectric or geothermal,
their operational conditions are such that they are not considered variable as defined in this report.
2
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 1. Summary of VRE Attributes Relevant to Capacity Expansion Models
VRE Attribute Physical Impact Relevance for CEMs
Variable VRE increases the variability of net load (load
minus VRE generation) because its available
power changes through time and space based
on changing weather patterns (e.g., wind
speed or solar irradiance)
Need for appropriate temporal
and spatial resolution to capture
variability and correlations with
other time-series variables
(e.g., load)
Uncertain VRE increases the uncertainty of net load
because the available power cannot be
perfectly forecast at all time horizons.
a
Requires methods to account for
adequate operating reserves
Near-zero
marginal cost
VRE resources have near-zero or zero variable
production costs because of negligible
operations, maintenance, and fuel costs
(relative to conventional technologies), and
when production-based subsidies exist, this
variable cost can be negative.
Requires proper accounting of high
fixed costs, zero variable cost, and
any relevant production- or
capacity-based subsidies;
potentially requires representation
of market operation impacts and
behavior
Lower capacity
value
As a consequence of VRE’s variability, VRE
resources have lower capacity value than most
conventional resources because of VRE’s
diurnal and seasonal patterns that may result
in low alignments of VRE generation with load
during times of highest system risk to reliability;
when resources are coincident with peak
demand, this contribution from VRE declines
with greater levels of VRE generation.
Requires appropriate methods to
account for VRE’s contribution
to resource adequacy needs using
time-synchronized load and VRE
data, and ideally is based on
probabilistic reliability approaches
for identifying highest risk periods
to reliability
Curtailment As a consequence of VRE’s variability, VRE
resources can experience times of curtailment
when the remainder of the generator fleet is
unable (for economic, reliability, or other
reasons) to further reduce its operating level to
accommodate VRE generation.
Consideration of VRE curtailment
using temporally-resolved, time-
synchronized load and VRE data
and key thermal generator operating
parameters such as minimum
generation level, ramping
constraints, and shut-down/start-up
costs
Geographically
dispersed
The dispersed nature of VRE resources
requires adequate transmission infrastructure
to transport electricity to end users.
Appropriate representation of
transmission network, potentially
including line flows and new
capacity enhancements, as well as
additional costs for “spur lines”
connecting dispersed VRE sites to
existing network infrastructure
Inverter-based VRE technologies are connected to the grid
through power electronic-based inverters, in
contrast to mechanically driven generators with
rotating mass that is synchronized to the grid;
inverters must be carefully designed to supply
necessary grid stability services.
Currently unknown, but might limit
“instantaneous” penetration of VRE
generators because of inertia
limitations
a
This uncertainty also impacts the capacity value and curtailment.
3
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
generators are dispatchable and have little variation in their performance across regions. Because
of their spatial and temporal diversity, VRE technologies typically require significant amounts of
data to be included in CEMs. These traits impose challenges to valuing VRE generation in
capacity expansion and dispatch models. Additionally, VRE technologies have evolved rapidly
in recent years, so their future costs are difficult to accurately predict (NREL 2017).
These challenges associated with representing VRE in CEMs motivated us to develop
a collaboration of four separate modeling teams to assess current best practices, share methods
and data, and identify future research needs
4
using the following four national-scale CEMs:
• Integrated Planning Model (IPM)—U.S. Environmental Protection Agency (EPA)
• National Energy Modeling System (NEMS)—U.S. Energy Information Administration (EIA)
• Regional Energy Deployment System (ReEDS)— National Renewable Energy Laboratory
(NREL)
• United States Regional Economy, Greenhouse Gas, and Energy (US-REGEN)—The Electric
Power Research Institute (EPRI)
This collaborative research project included two closely-related workshops that examined VRE
data and methodologies used by the four national-scale long-term planning models. The
workshops were held in December 2016 in Washington, D.C. and in June 2017 in Golden,
Colorado. Each of the workshops was one and a half days and covered a wide range of VRE
topics, including transmission and VRE resource data, VRE capacity value, dispatch and
operational modeling, distributed generation, and temporal and spatial resolution. The objectives
of the workshops were both to better understand these topics and to improve the representation
of VRE across the suite of models. And, each team incorporated model updates and performed
additional analyses between the first and second workshops. A summary of the additional
analyses and model “experiments” is provided in the appendix.
This report summarizes the findings and shares what we learned from the two workshops.
In particular, we emphasize the areas where there is still need for additional research and
development (R&D) on analysis tools to incorporate VRE into long-term planning and decision-
making. We note that other collaborative model improvement and comparison efforts have been
conducted in the integrated assessment modeling community, such as the Advanced Model
Development and Validation for the Improved Analysis of Costs and Impacts of Mitigation
Policies (ADVANCE)
5
project and the Energy Modeling Forum (EMF).
6
This report and
an accompanying paper (Mai et al. forthcoming)
7
both provide an overview of model
improvements and comparisons but applied to CEMs for the U.S. power sector.
4
Modeling teams from this collaboration also conducted a multi-model scenario analysis (Mai et al. forthcoming)
5
See a description of the ADVANCE initiative and a full set of publications at http://www.fp7-advance.eu/
6
See a description of the EMF initiative and a full set of publications at https://emf.stanford.edu/
7
Mai, Bistline, Sun et al. (forthcoming) presents a model scenario exploration, while this report focuses more on
methods and key factors for VRE representation.
4
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
This report is organized as follows. Section 2 provides a high-level summary of the four models
that were included in the workshops. Section 3 presents four themed summaries from the
workshop: spatial and temporal resolution, resource adequacy, economics of energy production,
and an “other” summary that includes ancillary findings from the workshops. These themed
summaries include model method improvements, sensitivity analyses, or both. Each summary
ends with a list of key R&D areas that apply to the theme. Finally, Section 4 presents the
conclusions from this work.
5
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
2 Model Summaries
Four CEMs were included in this comparison effort: IPM (EPA 2013, 2015b), NEMS (EIA
2017b), ReEDS (Eurek et al. 2016), and US-REGEN (EPRI 2017). These CEMs were developed
for different use cases. Because model development is largely guided by the intended use of the
model, the differences between the four CEMs is in part driven by their differences in end-use
application. As a summary of models and uses:
• EPA’s version of IPM is often used to evaluate various emission and environmental
policies, such as the Cross-State Air Pollution Rule (CSAPR) and the Mercury and Air
Toxics Standards (MATS).
8
As a result, EPA has focused considerable effort on the
representation of fossil-based generator technologies and associated emissions and
environmental impacts.
• ReEDS was developed primarily to analyze scenarios with high VRE penetration levels,
and it thus has a highly resolved VRE representation.
• NEMS is the EIA’s primary tool to provide projections for its Annual Energy Outlook
(AEO) reports, which provide a baseline examination of U.S. energy markets and facilitate
better understanding of future policies and market evolutions. NEMS has a full linkage of the
energy and economy sectors to allow it to more appropriately capture economic feedback and
foresight impacts into the energy sector evolution.
• US-REGEN is the newest of the four capacity expansion models. It was built to answer and
serve a diverse set of questions, interests, and stakeholders. It was designed to be flexible,
including customizable regions and time slices, an adjustable planning horizon, and the
capability to link its electric-only model with an economy-wide model, a unit commitment
model, and/or an end-use demand model.
Various characteristics of each of the four models covered in the multi-model workshops are
summarized in Table 2, Table 3, and Table 4, including some of the differences highlighted
above. Of particular focus in these tables are features relevant to the representation of VRE,
which was the focus of the workshops. The base version of each model is formulated as a linear
program (LP) to facilitate computational tractability.
9
Additionally, the four models assume a
central-planning framework that seeks to minimize the net present value (NPV) of the entire
system.
10
As a result, the models neither represent multiple decision makers nor capture explicit
choices of large actors (e.g., corporate power purchase agreements).
8
The IPM model version and projections presented in this report were developed by EPA’s Clean Air Markets
Division with technical support from ICF International, Inc. The IPM modeling platform is a product of ICF
Resources, LLC, an operating company of ICF International and is used in support of its public and private sector
clients.
9
LPs are used for optimal decision making in various industries and applications; as compared to the more complex
integer or non-linear mathematical programs, LPs can be solved far more quickly and directly with a guaranteed
single optimal solution. We note that, under certain scenario configurations, the US-REGEN model is run as a
quadratically constrained program or QCP.
10
Rooftop PV adoption is an exception—it is modeled using other methods that do not rely on minimizing system-
wide NPV.
6
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 2. Model and Computational Details
Model Institution Objective Computational
Requirements
Planning Horizon Foresight
IPM
U.S. Environmental
Protection Agency
(EPA) and ICF
Minimize the NPV of the
power sector's total
annual production costs
~10 hour run time on
computational server
Non-chronological, all
periods solved
simultaneously
Perfect foresight
NEMS
U.S. Energy Information
Administration (EIA)
Least cost optimization
for the U.S. electric
power sector; the EMM
projects capacity
planning, generation,
fuel use, transmission,
and pricing of electricity,
subject to inputs and
interactions with other
modules in NEMS.
~8-12 hour run time as
part of integrated
NEMS runs, ~4 GB
memory
NEMS solves annually
through 2050. In EMM
each solve-year
optimizes over a three-
period planning horizon
to examine costs over a
30-year period, which
consists of the current
year, the next year, and
the final 28 years of the
cost recovery period.
The model uses
convergent perfect
foresight within the
2050 planning horizon
by using prior run
results as input to the
current run. Out-of-
horizon years use
adaptive foresight.
ReEDS
National Renewable
Energy Laboratory
(NREL)
Minimize total system
cost using the 20-year
NPV
~8 hour run time, ~12
GB memory
2-year increments
through 2050
Foresight only for
natural gas and CO
2
prices
US-REGEN
Electric Power
Research Institute
(EPRI)
Maximize NPV of
surplus over the model
time horizon
(accounting for end
effects); minimize NPV
if electric sector only
model
Depends on spatial and
temporal resolution: ~1
hour run time, ~32 GB
memory for 48-state
runs
Customizable; for most
analyses, three-year
increments through
2030 and five-year
increments through
2050
Intertemporal perfect
foresight
7
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 3. Power Sector Constraints/Implementation
Model Temporal Resolution
a
Spatial Resolution
a
Transmission Representation Plant Retirements
IPM
12 time slices for each run
year (2 seasons x 6
segments) through 2030; 8
time slices (2 seasons x 4
segments) for all post-2030
run years
64 regions covering the
contiguous United States
(61 power market regions
and 3 power switching
regions), with 11 additional
provincial power market
regions representing
southern Canada
Firm and non-firm total transfer
capabilities between regions; joint
transmission capacity and energy
limits; wheeling charges between
regional transmission
organizations; static interregional
transmission losses
Economic retirements for all
non-VRE technologies; VRE
technologies assumed to
incur life extension costs to
continue operation indefinitely
NEMS
Hourly loads are divided into
3 seasonal periods (summer,
winter, and spring/fall). For
each season, the loads are
divided into 3 groups: peak
(highest 1%), intermediate
(next 49%), and base (lowest
50%), totaling 9 segments.
The generation of
electricity is accounted for
in 22 supply regions that
resemble the North
American Electric
Reliability Corporation
(NERC) reliability
assessment regions.
NERC collects data and 10-year
projections of demand, generating
and transmission capacity, and
capacity purchases and sales, by
region and/or by utility. These data
are used as input to capture firm
power transactions.
Announced retirements are a
model input. The model also
evaluates retirement
decisions for fossil and
nuclear based on whether
continuing operation costs
exceed revenues and if new
capacity is more economical.
ReEDS
17 Time slices (4 per day x 4
seasons + summer afternoon
super-peak) across one year
Contiguous United States
with 134 load balancing
areas and 18 resource
adequacy regions; some
representation of Canada
and Mexico
Approximate DC power flow
between 134 load regions;
susceptances and line capacities
updated between each solve
period; VRE spur lines represented
within regions
Age-based retirements for all
technologies; additionally,
minimum capacity factor-
based retirements for coal
US-REGEN
Customizable; typically 100+
"representative hours"
(Blanford et al. 2016) per year
Contiguous United States;
customizable regions
based on state boundaries
Pipeline representation of flows
across existing and new interstate
transmission lines; carrying
charges and line losses included;
$450 per kilowatt (kW)
transmission adder for new
renewables
Exogenous retirements for
most technologies due to
announced closures and age-
based retirements;
endogenous economic
retirements also possible
a
As applied to the dispatch process within the models
8
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 4. VRE-Specific Characteristics
Model VRE Spatial Resolution Out of Optimization
VRE Treatment
Distributed PV
IPM
Onshore wind resource is allocated to five potential
resource classes across three potential cost
classes for each of the 110 state and IPM region
combinations in the contiguous United States.
Solar photovoltaic (PV) resource is divided among
122 state and IPM region combinations.
None PV deployment reflected in net
energy for load taken from EIA’s
AEO. No distributed PV build
options are modeled in IPM.
NEMS
Solar: 22 regions
Wind: 22 regions and 4 resource classes
Between each solve period, a
statistical algorithm estimates
marginal curtailment rates using a
sequential 864 hourly model.
Small-scale solar PV is
endogenously modeled in the end-
use module of NEMS through a
hurdle model approach using Zip
code-level data.
ReEDS
PV: 134 regions and 9 resource classes
Concentrating solar power (CSP): 356 regions and
5 resource classes
Land-based wind: 356 regions and 10
resource classes
Offshore wind: 70 regions and 15 resource classes
Between each solve period, the
model calculates capacity value with
hourly data and estimates
curtailment and forecasting error
reserve requirement with a statistical
method.
Exogenous rooftop PV adoption
from dGen model
a
US-REGEN
PV: 5 resource classes for each model region; 3
solar technology options for each class (fixed tilt,
single-axis tracking, or double-axis tracking)
CSP: 5 resource classes for each model region
Onshore wind: 8 resource classes for each model
region
Offshore wind: 1 resource class for each
model region
None Distributed generation (including
rooftop PV) retail market model
iterates with the electric sector
model
a
See https://www.nrel.gov/analysis/dgen/.
9
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
3 Results and Discussion
3.1 Spatial and Temporal Resolution
One of the most apparent differences among—and significant challenges with—the models
included in this multi-model effort is model resolution, both in the spatial and temporal
dimensions. Because the level of spatial and temporal resolution used in these models is a
key factor in the representation of variability and correlation of load and VRE resources, the
resolution can have a substantial impact on both the VRE and overall model solutions. Too little
resolution can fail to capture important correlations or the full distribution of high- and low-
quality VRE resources and other time-series variables (see Table 1), while too much resolution
can cause the model to become computationally intractable or have data requirements that are
currently unrealistic. The “right” choice of resolution often depends on the research question
being addressed. And, because that choice is typically unknown, we explore in this section
the spatial and temporal resolution of the models and the effects of changing that resolution.
Spatial resolution is reflected by the number of subnational geographical regions, each of which
is typically associated with its aggregate load, VRE resource availability, and conventional
generator fleet. Depending on the model, these regions may be defined by state boundaries,
cross-state or sub-state balancing areas, or other planning-relevant zones. The intraregional
resolution of VRE resources varies by model and is a function of the granularity of VRE cost,
resource quality, and resource availability data. Each region is typically represented as a single
node within a transmission network topology.
11
As shown in Figure 1, the spatial extent of the
four models included here each captures, at a minimum, the contiguous United States, but IPM
also endogenously represents Canada.
12
Within this spatial extent, the models have a wide range
in the number and size of regions. This resolution sometimes varies by system component or
model configuration. For example, in ReEDS, some VRE resources are represented by 356
resource regions, while more spatially homogenous resources are resolved into the native 134
load balancing regions. US-REGEN provides a user-specified spatial resolution, up to the 48
contiguous states. Furthermore, the NEMS and IPM models have developed methodologies to
capture sub-regional boundaries that are either explicitly synchronized to the model regions or
implicitly embedded within the boundaries of smaller regional definitions, which are used to
capture state and regional policies within model constraints.
11
Each region in these models is a node, so 22–134 nodes are represented in them (see Table 3).
12
Other CEMs, including integrated assessment models and computable general equilibrium models, extend beyond
just the power sector and can have a much broader scope, covering other economic sectors and geographic regions,
such as entire continents or the full globe. As discussed earlier, NEMS includes a full representation of the U.S.
energy-economy, but our project is focused primarily on the electricity market module within NEMS.
10
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 1. Native spatial resolution of the four models: 64 IPM regions (top left), 22 NEMS
regions (top right), 134 ReEDS balancing area regions (bottom left), and 48 US-REGEN states
(bottom right)
Because of computational limitations, these four national-scale planning models are unable to
represent both the investment decisions and a fully resolved operational dispatch process across
a full year. As a result, representative time intervals (“time slices”) are typically used to represent
a full year of 8,760 hours (or a greater number of sub-hourly time steps). The temporal resolution
of the model dispatch refers to the number of such time slices. Numerous methods exist
exogenously selecting individual hours or representative time steps from a full year of data; these
include clustering techniques to select characteristic, aggregating similar hours, and using
characteristic time blocks by day and/or season (Getman et al. 2015; Blanford et al. 2016;
Nahmmacher et al. 2016; Santen et al. 2017). Temporal resolution is also reflected in the
underlying VRE data used in these down-scaling methods, as well as in metrics calculated
outside the optimization (but still endogenous to the model) to capture intra-time-slice VRE
characteristics. Aspects of the various temporal resolution and methods used by the four CEMs
are summarized in Table 5. One disadvantage of using time slices, which is shared by all these
models, is the non-chronological nature of the resulting dispatch steps, which augments
challenges associated with modeling energy storage, end-use demand response, and other
technologies with strong interdependencies across time slices. Furthermore, by nature of being
long-term planning models, the temporal extent of each of these models is decades into the
future, typically through 2050. However, the time steps along the investment time horizon
and model foresight can vary (see Table 2).
11
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 5. Summary of Temporal Resolution and Methods for the Four Models
Model Underlying
VRE Data
Resolution
Dispatch
Resolution
Dispatch Resolution
Method
Intra-time-slice VRE
Parameter Resolution
IPM
Hourly
(“8760”) load
and VRE time
series data
12 time slices
through 2030;
8 thereafter
6 stylized time blocks across
load duration curve for each
of 2 seasons; 4 stylized time
blocks for post-2030 run
years.
NA
NEMS
Hourly (8760)
load and VRE
time series
data
9 time slices 3 load-based time segments
(peak, shoulder, base) per
each of 3 seasons (summer,
winter, fall/spring)
Curtailment calculated
from 864 representative
hours (3 day types x 24
hours x 12 months)
(see Section 3.1.1)
ReEDS
Hourly (8760)
load and VRE
time series
data
17 time slices Fixed time blocks: 4 time
slices per day per each of
the 4 seasons, plus a super
peak summer afternoon
period
Capacity value
calculated from hourly
(8760) load and VRE
time series data (see
Section 3.1.1)
US-
REGEN
Hourly (8760)
load and VRE
time series
data
User-defined;
typically ~100
time slices
Clustering method to capture
both extreme events and
representative hours
throughout the year
NA
A key theme that emerged from the multi-model comparison workshops was that the selection
method for temporal and spatial resolution can be as important as the resolution itself. These
methods include techniques both outside and inside the optimization code. Outside methods
focus on capturing the subset of the most important locations or times in order to reduce model
complexity. Inside methods aim to improve resolution through dispatch decision variables, for
example, by increasing the number of time slices or increasing the number of VRE resource
classes. The outside methods include both methodologies that update parameters between solve
periods
13
and methodologies used to select exogenous parameters (e.g., how should time slices
be created?). The way in which these outside methods can be pursued range from direct
incorporation (e.g., building a methodology into a model) to using a separate, higher-
resolution model to inform parameter selection.
Much of the current model enhancement work among the teams included in this multi-model
exercise focuses on developing sophisticated outside methods. This effort includes representing
temporal and/or spatial system interactions that would be too complex and/or computationally
burdensome to include within the model while maintaining an LP formulation. For several
of the modeling teams, improved temporal methods aim to calculate metrics outside the
optimization to approximate intra-time-period variability and/or uncertainty. We detail two
examples below, from the NEMS and ReEDS models. Improved spatial methods include the
13
NEMS and ReEDS are sequential models, so parameters can be updated between solve periods. For example,
after solving for the year 2020, ReEDS and NEMS employs nonlinear VRE calculations to update VRE parameters
that impact the value of VRE (e.g., from curtailment) before proceeding to the next year.
12
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
use of higher-resolution exogenous supply or cost curves to approximate intraregional
distributions of site-level features for VRE and other resource availability. The resolution can
be further improved by subdividing these resources into resource quality classes. In the spatial
method examples below with the IPM, US-REGEN, and ReEDS models, we show how the
seemingly simple choice of the number of regions can yield very different model results based
on the treatment of VRE resources.
3.1.1 Examples of Temporal Resolution Methods
The “outside” temporal resolution selection methods detailed here reflect efforts by the NEMS
and ReEDS modeling teams to improve the calculation of intra-time slice metrics, specifically to
capture the impact on PV curtailment and VRE capacity value respectively. The NEMS example
demonstrates the impact of increasing the number of model time slices. As we will see, there are
tradeoffs in computation time and model results between the numbers of time slices and how
intra-time-slice characteristics are represented. The results presented here suggest that improved
methodologies for calculating intra-time-slice characteristics provide a similar temporal
resolution benefit to approaches that only increase the number of time slices but they do so
with a smaller increase in computational burden.
NEMS represents load, VRE resources, and dispatch across nine time slices: three seasons
(summer, winter, and fall/spring) each with three load-based time segments (peak, shoulder, and
base). To capture intra-time-slice behavior, EIA implemented a “864 method” to estimate PV
curtailment within these time slices based on 864 hours across the year (12 months each with 24
hours and 3 load-based day types that correspond to weekday and weekend variations in
demand). Within each of the 864 time slices, the method computes the adjusted energy value of
PV after any generation in excess of net demand (after accounting for minimum generation limits
of units with limited cycling capability). As solar generation begins to saturate the demand
during daylight hours within each month, the apparent value of the resource (as measured by the
cost of displaced energy) is reduced. This energy value parameter is then aggregated back to the
nine time slices used in the LP, and it is passed to that process as an adjustment to the cost of PV.
In addition, this approach has been extended to wind curtailments and energy storage.
While developing the new approach, EIA also investigated increasing the temporal resolution of
the LP itself, focusing on adding time slices to provide more resolution around daylight and
nighttime hours. While time slices were added as a way to quickly assess the impact of the low
temporal resolution on PV model results prior to more-involved model development, the
expectation was that the impact on execution time would be too severe to operationalize in the
production NEMS model. However, this limited development provides a reasonable benchmark
against which the final 864 algorithm can be compared for both model impact and performance.
A comparison of with (“w/ Curt”) and without (“No Curt”) curtailment using the 864 method
and 9 versus 15 time slices is shown in Figure 2. Deployment is generally unaffected, except for
PV, which sees a reduction in buildout with both the 864 method and 15 time slices, suggesting
both methods achieve similar temporal resolution improvements of capturing the curtailment
impact at higher PV penetration levels.
14
However, increasing the number of time slices had a
significantly greater impact on run time than the 864 method. In high VRE deployment
14
Because of the wind/PV trade-off, wind deployment is increased as the PV deployment decreases.
13
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
scenarios, run times increased by no more than 6% with the 864 method, while run times
increased by over 20% with 15 time slices. In summary, the 864 curtailment algorithm leads to
similar deployment results, represents PV generation with higher granularity, and has minimal
impact on run times, compared to increasing number of time slices.
Figure 2. NEMS deployment results with (“w/ Curt”) and without (“No Curt”) 864 Curtailment
method using 9 versus 15 time slices
HIVG = high variable generation
The second example of “outside” methods to improve temporal resolution is with the ReEDS
“8760 method” that is based on time-synchronous hourly load (i.e., 8,760 hours) and VRE data.
As shown in Figure 3, this approach approximates the capacity value (CV) of VRE as the
difference between the top 100 hours of the load duration curve (LDC) and the net load duration
curve (NLDC). This approach further estimates the marginal CV of potential new VRE
generators as the difference between the NLDC and incremental NLDC (NLDC(δ)). This
methodology is detailed by Frew et al. (2017) and was developed in coordination with a similar
methodology implemented in NREL’s Resource Planning Model or RPM (Hale, Stoll, and
Mai 2016). While this LDC-based approach for CV is not itself novel (e.g., IEA 2015), it
(1) demonstrates how a simplified 8760 hourly method more accurately captures CV trends
than a statistical method within the ReEDS CEM and (2) provides a flexible modeling
framework from which other 8760-based system elements (e.g., demand response, storage,
and transmission) can be added to further capture important dynamic interactions, such
as curtailment.
14
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 3. ReEDS LDC-based approach to calculating CV
ReEDS previously estimated CV using a statistical approach that considered simple summary
metrics (variance and expected value) from the underlying hourly load and resource data within
each of the 17 time slices.
15
ReEDS optimizes investment decisions within two-year solve
periods, sequentially solving from the present day system out to the model horizon of 2050.
The CV parameters are updated between each of these two-year solve periods and then used
in the subsequent solve period in ReEDS to quantify each VRE resource’s capacity contribution
to the planning reserve constraint.
Results suggest that the new 8760 method offers a more accurate representation of VRE CV in
ReEDS than the former statistical approximation method with less than a 10% increase in solve
time. The marginal PV CV outputs derived by the former ReEDS statistical method and the new
8760 method are shown in Figure 4. The new ReEDS method better captures the declining
capacity value of VRE with increasing levels of penetration (see also Section 3.2). Previous work
has shown, and Figure 4 supports, that the former ReEDS CV method yields abrupt changes in
CV between the different time slices, particularly between summer afternoon and evening
(Sigrin et al. 2014). These results can be seen in Figure 4’s left pane by the sharp drop in the
former ReEDS method marginal CV around the 7% PV penetration level, where the planning
reserve constraint binding time slice shifts from summer afternoon to evening (yellow
diamonds). Furthermore, as the right pane shows, the coarse time-slice-based values in the
former ReEDS method often estimate persistently high CVs for PV even at relatively high
penetration levels. The new 8760-based method (red triangles), which looks across the top 100
net load hours to calculate an annual CV, results in a smoother and more rapid decline in CV.
15
ReEDS still uses this statistical approach for estimating VRE curtailment and forecast reserve errors. And,
as with the CV method, these other statistical approaches happen “outside” the optimization.
15
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 4. Incremental PV CV using the former and new ReEDS CV method in the Austin, Texas,
region (left) and Southern California region (right)
In summary, these examples show how the temporal resolution can significantly impact the value
of VRE to the system. Insufficient temporal resolution can lead to overvaluing VRE, especially
PV, by underestimating VRE curtailment and overestimating capacity value. However, increased
temporal resolution typically requires higher-resolution data and often increases model runtime.
3.1.2 Examples of Spatial Resolution Methods
The four models discussed in this report differ significantly in the number of geographic
regions they represent. In this section, we show how the method for constructing various spatial
resolutions can yield opposing trends in some model results. In particular, as shown with
examples from the IPM, US-REGEN, and ReEDS models, the treatment of VRE resources can
drive total system costs to either increase or decrease with greater spatial resolution, depending
on the method for increasing the resolution.
The IPM model includes 64 U.S. regions. To examine how the spatial resolution of transmission
system representation impacts modeling projections, an alternate version of IPM was created by
eliminating transmission constraints, charges, and losses between select IPM regions. The
aggregated transmission regions were designed to approximate the 22 regions of the EMM in
NEMS.
16
All other regional parameters, such as resource supply curves, were maintained at the
64-region level. To quantify the impact of this change, a minimum generation constraint was
imposed in both versions of IPM that required wind and solar resources to supply generation
equal to 40% of load by 2040.
17
In the version of IPM with aggregated transmission regions, the
model was able to produce the prescribed level of VRE penetration in 2040 at a 5% savings in
total annual production costs ($9.5 billion in 2016$), relative to the standard IPM version. These
savings were achieved through greater utilization of more remote but higher-resource wind,
18
which allows for (1) 13 GW in avoided solar capacity, (2) a slight shift away from higher fixed-
cost coal to more flexible natural gas resources (an incremental 9 GW of coal is retired under the
16
Transmission limits, charges, and losses are maintained between the aggregated regions.
17
The constraint requires VRE generation equal to or greater than 17.5% of load by 2025; 22% of load by 2028;
25% of load by 2030; 32.5% of load by 2035; 40% of load by 2040.
18
The wind fleet is 1 GW smaller but produces 20 terawatt-hours (TWh) more generation in 2040.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0% 10% 20% 30%
Marginal PV CV
PV Energy Penetration
New ReEDS - annual
CV
Former ReEDS -
binding timeslice
0
0.05
0.1
0.15
0.2
0.25
0.3
0% 20% 40%
Marginal PV CV
PV Energy Penetration
New ReEDS -
annual CV
Former ReEDS -
binding timeslice
16
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
aggregated transmission region version), and (3) a small overall reduction in total capacity and
generation through avoided transmission losses. These factors are reflected in the components
of production cost savings presented for a sample year in Figure 5, with the primary driver of
cost savings in the aggregated transmission region run being the avoided 13 GW of new solar
capacity. On an NPV basis, the cost savings from 2016 to 2040 realized by aggregating
transmission regions is $104.4 billion (2016$), of which $73.0 billion is attributable to avoided
capital expenditures. In other words, if IPM lacked the regional resolution in VRE-relevant
constraints that was eliminated for this scenario, it might underestimate the system cost and
capacity additions needed to reach this level of VRE generation.
Figure 5. Annual production cost savings from 22 aggregated transmission regions
(versus 64 native regions) for sample year 2040
US-REGEN performed a similar comparison of the 48 contiguous states against a smaller set
of 15 regions, as detailed in Bistline et al. (2017). The 48-state resolution is currently the finest
spatial resolution for planning decisions in US-REGEN. Similarly to the IPM example, the
intraregional transmission constraints were collapsed within the 15-region case, but other
intraregional parameters were maintained. Like IPM, the US-REGEN model with lower spatial
resolution had lower system costs (see Figure 6). Cost differences are amplified by policy
constraints such as a 50% national renewable mandate by 2050 (the “RPS” scenario in Figure 6)
as well as market conditions such as gas prices or limitations on trade. Incremental electric sector
costs relative to the reference in Figure 6 are shown assuming full interregional renewable
energy certificate trade and with restrictions on such trade. With fewer regions, there were fewer
transmission constraints, making it easier to access lower-cost VRE resources. Total system costs
were approximately 2% lower in the 15-region case than in the 48-region case.
-
2.0
4.0
6.0
8.0
10.0
Billion 2016$
Variable O&M
Fixed O&M
Fuel
Capital
0.3
0.6
1.6
7.0
17
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 6. Incremental compliance cost of a 50% wind and solar RPS (relative to a reference
scenario without these mandates) with 15 versus 48 US-REGEN regions (Bistline et al. 2017)
In the ReEDS example, instead of adjusting transmission constraints to match varying levels of
regional aggregation, the impact of spatially aggregating VRE resources within a constant
transmission topology was evaluated (Krishnan and Cole 2016). Three aggregation scenarios
were considered: 134 native regions (“REF”), 48 contiguous U.S. states (“STATE”), and 13
approximate North American Electric Reliability Corporation regions (“NERC”). For each level
of aggregation, the VRE cost and performance characteristics were averaged (capacity-weighted)
to obtain a single supply curve for each resource within each region evaluated, but other regional
parameters, such as the transmission system were left at the native 134-region resolution.
19
Unlike the trend observed by IPM and US-REGEN, the total system cost increased as the VRE
resource was aggregated, as shown in Table 6. This is driven by the impact of averaging higher-
and lower-quality VRE resources, yielding smoother supply curves that fail to capture the full
distribution of resource quality and resulting competitiveness among sites and across
technologies. With a more aggregate resource representation, the model can no longer see the
best-quality, lowest-cost sites that were available in the more spatially resolved representation.
19
Within each region, ReEDS includes 10 wind classes (called techno-resource groups), 9 PV classes, and 5 CSP
classes. In this example, the classes were maintained but were aggregated with other similar classes in that region.
18
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 6. ReEDS’ Total System Cost with 134 (REF), 48 (STATE), and 13 (NERC) Regions
for VRE Representation (Krishnan and Cole 2016)
Category (2015$) REF STATE NERC
Conventional capital 345 356 356
Conventional O&M 840 844 846
Conventional Fuel 2,160 2,190 2,235
Renewable capital 614 598 570
Renewable O&M 239 236 237
Renewable Fuel 24 24 25
Storage capital 2.5 2.2 2.1
Storage O&M 9.3 9.1 9.1
All transmission 61 63 66
Water 0.02 0.02 0.02
Total 4,298 4,323 4,347
Difference 25 49
In summary, these examples show that insufficient spatial resolution can lead to either under- or
over-representing the cost of building out a system, depending on how the spatial aggregation
takes place. Aggregating the transmission leads to lower costs because lower-cost but more-
remote resources can contribute to the system with fewer transmission limitations. Aggregating
the VRE resources leads to higher costs because the highest-quality resource sites are averaged
with mid- or low-quality sites, reducing the model’s ability to “see” the lowest-cost VRE
potential. Higher spatial resolution, however, necessitates higher-resolution spatial data and,
because of the linear nature of these models, can lead to false precision (e.g., a very minor cost
difference between two small regions might mean all new capacity is built in the slightly less
expensive region). Additionally, increases in spatial resolution can increase model run time.
3.1.3 Recommendations for Future Modeling R&D
Developing improved temporal and spatial resolution methods is an active area of research
among the four modeling teams included here. Moving forward, the teams have identified
several areas of priority for future R&D, including:
• More systematic study of the cost/benefit tradeoffs of increased resolution—The general
understanding is that higher resolution leads to improved model results but also incurs
additional computational burden. That tradeoff strongly depends on the methods and model
used. More work is needed to better quantify and categorize the impact of both the resolution
and method on model accuracy and computational cost. This necessarily requires developing
and testing input data with greater resolution and/or scope, such as sub-hourly load and VRE
time series data and/or multiple years of such time series data. Such experiments can help
modelers better prioritize resolution tradeoffs and understand the scenario assumptions under
which greater model resolutions matter most.
19
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
• Understanding of how much resolution is needed for VRE deployment projections—VRE
characteristics are generally better represented at higher resolution, but the marginal value of
additional resolution will decrease as the resolution increases. The value of the additional
resolution will also depend on the level of VRE penetration (e.g., a 100% VRE study would
need more resolution than a 50% VRE study) A quantitative understanding of how much
resolution is needed in making VRE deployment projections would allow modeling teams to
ensure they have a sufficiently robust representation of VRE characteristics.
• Metrics to quantify if and how much results improve with increased resolution or selection
methodology—In order to assess the cost/benefit tradeoffs of the resolution and method
choices, metrics need to be developed to provide equitable comparison and benchmarking of
these choices and the model used. These standards would ideally also serve to validate new
or improved methods. The metrics can also be used to inform decisions as to whether
methods should be directly incorporated within a model or if methods should be used to
parameterize existing model formulations.
• Methods that can capture chronology for representing energy storage, unit commitment, and
dispatch—The modeling techniques described here generally ignore many chronological
issues that are important for a number of operational constraints. There is a need for methods
that either can incorporate chronology directly into the models or can parameterize the
chronological aspects in a way that the current methodologies can represent those
chronological aspects.
3.2 Resource Adequacy
One of the key factors that long-term models must consider is whether there is sufficient capacity
to maintain system reliability. Resource adequacy refers to the need to have enough available
resources to meet anticipated demand while accounting for a reasonable number of
contingencies. Today’s utilities and regulatory bodies typically measure resource adequacy
using a planning reserve margin. Details of planning reserve margin calculations and definitions
vary by organization, but in general the planning reserve margin represents the ratio of available
capacity divided by the expected peak demand, and it is expressed as a percentage value over
100%. For example, if a region expects to have 12 GW of available capacity and has a 10-GW
peak, the planning reserve margin would be 20%. Reference planning reserve levels range from
10% to 20% (NERC 2017) and are typically designed around a loss-of-load probability, such as
one day of lost load per 10 years of operation. Nearly all regions in the United States and Canada
currently have a higher planning reserve margin than the NERC reference levels (see Figure 7).
The actual process of determining an appropriate planning reserve margin for a region includes
a mix of technical analysis, economic analysis, and regulatory interactions.
20
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 7. Reserve margins for each NERC assessment area (NERC 2017)
Capacity expansion models face two primary difficulties when representing resource adequacy.
The first difficulty lies in determining the availability of VRE resources for meeting peak
demand, or in other words, the contribution of a VRE generator toward the planning reserve
margin.
20
For example, how much can system operators rely on a wind or PV plant to be
producing energy during peak demand periods? This contribution has typically been referred
to as capacity value or capacity credit. The second difficulty lies in determining the level of
resource adequacy that is sufficient. As shown in Figure 7, current reserve margins are generally
higher (and sometimes much higher) than the reference reserve margin. Requiring too much
capacity can result in a more expensive, overbuilt system, while too little capacity can result in
expensive loss-of-load events.
21
We discuss these challenges in detail below.
3.2.1 VRE Contributions to Resource Adequacy
Several methods have been developed to try to address the first difficulty of representing VRE
generators’ contribution toward resource adequacy. The methods used by the four capacity
expansion models considered in this work are summarized in Table 7. Three of the four models
assess contributions of both conventional and VRE capacity toward the NERC reference reserve
margins, while US-REGEN uses a different methodology that derates the conventional capacity
based on a calculated availability factor (see EPRI 2017, Section 2.3.3) and applies that derated
capacity to the peak demand. Specifically for VRE capacity contributions, all of the models rely
on more detailed underlying data, such as hourly wind or solar data. These data are then
summarized outside the model optimization algorithm (see Section 3.1) and incorporated into
the optimization in order to allow the model to make appropriate investment decisions that will
result in systems with sufficient capacity. Though the algorithms and data requirements differ
20
Conventional generators have long been a major part of the power system and do not rely on variable resources
such as wind or sunlight, so their contribution toward resource adequacy is generally well understood and more
easily represented.
21
The CEMs in this work do not model loss-of-load events. Rather, they require a certain amount of resource
adequacy and assume the level of resource adequacy is sufficient to meet required reliability levels. For this reason,
it is important that the resource adequacy levels are appropriately represented in CEMs.
21
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
among models, they generally calculate the anticipated generation from VRE resources during
periods of peak net demand.
22
Table 7. Summary of How Conventional and VRE Capacity Contribute to Resource Adequacy
for the Four Models
Model Resource Adequacy
Requirement
Conventional Capacity
Contribution to
Resource Adequacy
VRE Capacity
Contribution to
Resource Adequacy
VRE Data Used
for Resource
Adequacy
Calculations
IPM
Planning reserve
margin based on
NERC reference
levels
Installed net
summer capacity
Specified exogenously
based on NEMS
capacity value outputs
scaled by difference
between IPM and
NEMS capacity factors
Seasonal
generation
profiles
NEMS
Planning reserve
margin based on
NERC reference
levels
Installed net
summer capacity
Statistical method that
accounts for spatial and
temporal correlation of
wind/solar resources
within a region
Average daily
generation by
month profiles
ReEDS
Planning reserve
margin based on
NERC reference
levels
Installed net
summer capacity
Average capacity factor
during the highest 100
net load hours
Hourly data from
2006 for load and
PV; hourly typical
meteorological
year data for
wind
US-
REGEN
No reserve margin—
capacity derated
to account for
contingencies
Installed net summer
capacity derated using
availability factors
Capacity factor during
time slice with peak
net load
Hourly data from
2010 for load,
PV, and wind
The lack of a metric for assessing whether a model is getting things “right” is one of the
challenges of capturing VRE contributions. The general trends for the capacity value of VRE
are well understood, namely, that wind has a low capacity value that does not change
dramatically with penetration and that PV begins with a high capacity value that declines rapidly
with penetration. The capacity value trends can be compared across models (see Figure 8) to
ensure they are well behaved. In addition the capacity expansion model results have often been
checked against more detailed models to ensure the systems they project do not drop load (Mai et
al. 2012; Jorgensen, Mai, and Brinkman 2017; Bistline 2017), but there is no way to conclusively
prove that the systems will meet the reliability targets anticipated by the reserve margins or
availability factors. The comparison against the more detailed models, however, can reveal
shortcomings of methodologies that can lead to improved methodologies. As a result of these
challenges, modeling VRE contribution toward resource adequacy is still an active area of
research (Frew et al. 2017), and all four modeling teams have identified areas in which they wish
to improve their VRE representation for resource adequacy.
22
Net demand is total demand minus VRE generation.
22
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 8. Capacity value for wind (left) and PV (right) as a function of PV and wind penetration
“Other studies” are from Holttinen et al. (2016) for wind and Mills and Wiser (2012), Denholm and Hummon (2012),
and Jorgenson, Denholm, and Mehos (2014) for PV
One specific challenge for representing the capacity value of VRE in long-term planning models
is data (Gami, Sioshansi, and Denholm 2017). Planning models are projecting new builds of
VRE generators, and thus, no historical data exist for how well the generation from these new
builds will align with peak demand. Furthermore, because weather patterns change from year to
year, it is not clear how multiple years of data should be used to determine VRE capacity value.
For example, if a wind generator has a low output in year one during the peak hour and a high
output in year two during the peak hour, should the generator have a low value, high value, or
some combination of the two values for its capacity value contribution? Because of processing
time and computing resource limitations, CEMs tend to look at a deterministic year for load and
VRE availability rather than use a probabilistic approach or one that considers multiple years-
worth of data. Additional years of high-resolution data are available for wind and PV, but lack of
high-resolution load data has been a shortcoming that has made it difficult to expand the models
to use multiple years of data.
3.2.2 Determination of the Resource Adequacy Level
Although VRE treatment for resource adequacy has received significant attention, significantly
less research has been directed toward determining the appropriate resource adequacy level
of future power systems. Reference case or business-as-usual scenarios in capacity expansion
models regularly include increasing shares of VRE (Cole et al. 2016; EIA 2017a; Bistline et al.
2017). And, the models are often employed to analyze scenarios with significant changes, such
as carbon reduction scenarios (EPA 2015a), rapid electrification (Steinberg et al. 2017), alternate
oil and gas resource and technology projections (EIA 2017a), and deep decarbonization
scenarios (Bistline and Chesnaye 2017). The systems projected in even the reference scenarios
can be substantially different from today’s system. Because the models use current or historical
data to determine resource adequacy (based on NERC’s reference-level reserve margins or on
historical plant availability), it is not clear that these metrics will still be sufficient for future
power systems that in some cases are vastly different from today’s system.
0%
10%
20%
30%
40%
50%
60%
70%
80%
0% 10% 20% 30% 40% 50%
Wind Capacity Value
Annual Energy from Wind
ReEDS
REGEN
NEMS
Other studies
0%
10%
20%
30%
40%
50%
60%
70%
80%
0% 10% 20% 30% 40% 50%
PV Capacity Value
Annual Energy from PV
ReEDS
REGEN
NEMS
Other Studies
23
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
3.2.3 Other Resource Adequacy Challenges
In addition to the challenges laid out above, there are other important factors for addressing
resource adequacy in long-term planning models. As can be seen in Table 7, the models have
different methods for treating conventional capacity, and these differences can end up being
significant when aggregated over the entire U.S. power system. Moreover, the primary drivers
for the need for new capacity are retirements of existing generators and growth in peak demand.
The four models have very different methods for treating existing plant retirements, which
results in vastly different needs for new capacity (see Figure 9). Peak demand in the four models
is typically scaled according to projected electricity demand growth, where all four models rely
on the integrated NEMS model to project how annual electricity demand will change over time.
Additionally, changes in end-use demands are challenging to capture, especially with respect to
how they might alter load change and impact peak demand requirements. In many analyses, load
shapes are assumed to be constant over time.
23
All these factors influence how much capacity is
needed and therefore the relative economics of new VRE generators.
Figure 9. Coal capacity (left) and nuclear capacity (right) in the model’s reference case scenarios
Finally, although all four models allow capacity to be shared between regions, little work has
been done to understand the load diversity between regions in order to assess the benefits and
limitations of sharing power between regions. This factor becomes especially important when
considering very long-distance transmission lines that might connect regions with different
time zones and therefore very different load shapes.
23
NEMS includes a demand-side module, so the extent to which load shapes are updated over time depend
on the assumptions and methods included in this module.
24
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
3.2.4 Recommendations for Future Modeling R&D
Although significant progress has been made for representing resource adequacy, there is still
substantial need for increased understanding and methodologies. The key R&D
recommendations we identified with respect to resource adequacy include:
• Understand how accurate VRE capacity value estimates need to be—Capacity value of VRE
is often discussed as an important element to capture in long-term planning models, but the
impact of underestimating or overestimating the capacity value has not been quantified.
Relatedly, the methodologies used by planning models for VRE capacity value estimates are
similar but vary considerably in their data and computational requirements, and to our
knowledge, no one has performed a comprehensive comparison and validation of the data
and techniques to understand the various tradeoffs of the methodologies.
• Understand how much capacity is needed, especially for higher-penetration VRE futures—
Models typically rely on NERC reference reserve margins or historical data to specify
required capacity needs, but the current power system exceeds those recommended levels.
Improved understanding of historical reserve margin levels and what drives those levels can
inform how resource adequacy requirements are made for long-term planning scenarios.
Additionally, there might be value in incorporating more fundamental reliability metrics
(e.g., loss-of-load probabilities) within the planning models.
• Improve VRE and load time series data—Models typically rely on a single year of simulated
data that might not capture inter-annual weather patterns, load shapes, or less-common
events. Multi-year data sets could potentially lead to resource adequacy estimates that are
more robust. Additionally, validating or improving simulated data based on historical
measured data can reduce or remove shortcomings associated with using simulated data
for long-term projections.
• Improve data and methods for estimating retirements—Retirements represent a major driver
for new capacity needs, but they are challenging to estimate within models. Cost data for
current generators is difficult to obtain, and therefore methodologies are often built to
accommodate the limited data that are available. Methodologies that can apply non-cost data,
such as utilization or contract data, might effectively supplement the limited cost data.
Any improvements to the accuracy or coverage of existing cost data sets will improve
retirement estimates.
3.3 Economics of Energy Production
Because capacity expansion models are typically least-cost optimization models (see Table 2),
the relative economics of the various technologies are a primary driver of model results. Because
of current policies, such as renewable portfolio standards and the sometimes long lead-time to
build new plants, the economics play a lesser role in the short-term
24
but are a more significant
driver over the long term.
24
Economics do impact near-term decisions, but in many cases, the decision to build a new plant in the near term
has already been made and the economics inside the model therefore do not impact whether that plant comes online.
25
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
At low penetration, nearly all VRE generation can be utilized. However, as the VRE penetration
increases, the potential for curtailment, which is the intentional reduction in generation from
VRE generators when generation exceeds load, can substantially increase (Denholm and
Margolis 2016; Denholm, Clark, and O’Connell 2016). VRE curtailment can result from
insufficient transmission capacity to export surplus power to neighboring regions, the inability
to store surplus energy, and/or the inability to ramp down generation from committed thermal
units (Bird, Cochran, and Wang 2014; Fink et al. 2009).
The specific representation of VRE curtailment varies across models, but the methods are all
essentially doing an accounting to determine the amount of VRE generation that cannot be
absorbed by the system. Assumptions with respect to transmission limitations, storage, and
thermal unit minimum generation levels drive the ability of the system to absorb more VRE,
and those assumptions certainly vary from one model to another. Comparing the curtailment
outputs among models can be especially challenging because actual curtailment rates tend to be
low (<5%) even at fairly significant penetration levels. Instead, curtailment calculations within
capacity expansion models are usually compared against curtailment estimates from an hourly
chronological model. For example, DOE (2012), Mai et al. (2012), and Jorgensen, Mai, and
Brinkman (2017) compared ReEDS curtailment estimates against those of an hourly production
cost model. And, these comparisons have shown that the current methodologies capture
curtailment fairly well for the specific scenarios that were examined. However, it is unclear
whether these methodologies will remain robust across a wider range of scenarios, such as those
with significant amounts of inflexible load, restricted transmission expansion, or high
penetrations of new storage capacity. Additionally, the impacts of unit commitment and ramping
restrictions are typically ignored in the capacity expansion models, and it is unclear when or how
frequently their exclusion can have an impact on model results.
If VRE curtailment is not represented in a capacity expansion model, that model is likely to
overestimate the value of VRE technologies. Given that today’s VRE penetration levels are still
relatively low, the exclusion of curtailment is unlikely to significantly impact near-term
modeling results.
25
However, because VRE curtailment increases as VRE penetration increases,
medium- and long-term projections are likely to be significantly impacted. Solar PV economics
are especially impacted by curtailment because PV production is typically more strongly
autocorrelated than production from wind plants, which results in a steep increase in curtailment
rate as PV penetration increases (Denholm and Margolis 2016). Figure 10 shows the impact of
not including curtailment within the NEMS model reference scenario for this exercise. Excluding
curtailment leads to increased solar penetration as well decreased gas, wind, and coal generation.
Without the curtailment represented in the model, utility-scale PV generation nearly doubles.
This scenario is approaching 20% VRE penetration, so the impact of curtailment in higher
penetration scenarios would be even more significant.
25
Regionally, the exclusion of curtailment might be impactful. For example, California experienced solar
curtailments in the first half of 2017, and ERCOT had high levels of wind curtailment before expanding
transmission and implementing new operational practices.
26
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Figure 10. NEMS generation mix in the reference case when curtailment is included (left)
and the difference in generation when curtailment is not included (right)
Operating reserves needed for VRE integration are another aspect of VRE representation that can
impact economics. Higher penetrations of VRE typically require a greater amount of operating
reserves to account for short-term errors between VRE forecasted and actual generation. Sample
scenarios using the NEMS model demonstrated this through adjusting the level of spinning
reserve capacity required as a percentage of available operating VRE capacity in each timeslice.
For low operating reserve requirements (where the VRE-specific spinning reserves contribution
was at 25% of VRE available capacity or less),
26
the impact on the capacity expansion solution
was negligible. At higher requirements (where the VRE-specific reserves were at 50% of VRE
available capacity or more), the impact became more pronounced.
Another major component of the economic competitiveness of VRE is the way in which
financing is treated in the model. VRE technologies are capital-intensive technologies, meaning
nearly all their cost is upfront capital cost associated with their construction and that their O& M
costs are relatively low. In contrast, fuel-burning technologies such as coal, natural gas, and
biopower plants can have significant fuel costs that are spread over the lifetime of the plant.
Because of the capital-intensive nature of VRE technologies, they are more strongly impacted by
financing assumptions; lower financing costs can make them more competitive with fuel-burning
generators, while higher financing costs can do the opposite. We found a range of financing
assumptions across the four models (see Table 8), and we also found that measuring the quality
of financing assumptions can be especially difficult, especially when considering that projections
extend to 2050 or later.
26
This operating reserve requirement is the fraction of the available VRE capacity in each time slice that must
be matched with spinning reserves.
0
1,000
2,000
3,000
4,000
5,000
6,000
2010 2020 2030 2040 2050
Generation (TWh)
With Curtailment
Utility PV End-use PV Other RE Coal
Petroleum Natural Gas Nuclear Other
-200
-150
-100
-50
0
50
100
150
200
2010 2020 2030 2040 2050
Difference in Generation from Figure at Left
(TWh)
Difference without Curtailment
27
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Table 8. Financing Assumptions Used in the Four Models
IPM:
Merchant
IPM:
Utility
IPM:
Overall
NEMS
(AEO2017)
a
US-
REGEN
ReEDS
Debt Interest Rate
5.30% 2.50% — 3.7% — 5.4%
Rate of Return on Equity
10.10% 5.30% — 8.2% — 10.2%
WACC
5.90% 3.01% 3.88% 5.1% 5% 5.4%
Economic Evaluation
Period (years)
20–40 20–40 20–40 30 25–100 20
All percentage values are in real (as opposed to nominal) terms.
a
Financing in NEMS varies over time; the values included here are the long-term financing values in
the model.
The impact of varying financing assumptions on the levelized cost of electricity (LCOE) for
a wind plant is shown in Figure 11. The four WACC levels from Table 8 are shown, along with a
range of cost recovery periods from 20 years to 30 years. There is a difference of about $6/MWh
across the WACC values (holding economic lifetime constant) and across economic lifetime
values (holding WACC constant).
27
A sample scenario using the ReEDS model showed that
changing the economic lifetime of all generators from 20 years to 30 years led to a 33% increase
in VRE capacity under reference scenario conditions.
Figure 11. Impact of financing assumptions on the LCOE of a wind plant
LCOE calculated using the 2016 Annual Technology Baseline (ATB) spreadsheet (NREL 2016).
27
LCOE values are calculated using the 2016 Annual Technology Baseline spreadsheet (NREL 2016).
28
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
The cost of connecting VRE resources to the grid is the final aspect of the economics of VRE
that we considered in our CEM modeling workshops. VRE generators are often built in locations
that are more remote and thus require some investment in the transmission system to move the
VRE generation to a load center. This area of representing VRE transmission connection costs
has received much less attention in the literature, with most of the published work relegated to
model documentation. The methods used to represent these costs for the four models are
summarized in Table 9. Interconnection values of $100/kW–200/kW represent approximately
5%–10% of the cost of a wind plant and approximately 5%–20% of the cost of a utility PV plant,
so these values are not insignificant. Comparison of methods with actual historical builds and
improved methods for identifying new VRE sites could lead to more robust methods and more
confidence in transmission connection costs.
Table 9. Summary of Transmission Connection Costs for VRE Technologies
IPM
a
NEMS
a
ReEDS US-REGEN
Type of Cost
Multiplier Multiplier Adder Adder
Type of Values
Supply curve Supply curve Supply curve Fixed value
Technologies Included
Wind Wind Wind, PV, CSP Wind, PV
Typical Value
~$200/kW ~$200/kW ~$100/kW $200/kW
Typical Range of Values
0–$2,000/kW 0–$2,000/kW $10–$1,000/kW $200/kW
a
IPM and NEMS do not apply a separate interconnection value to new wind capacity but instead uses a
range of multipliers to capture cost increases stemming from a variety of factors, including distance from
existing transmission, site accessibility challenges, population proximity, and other factors. The capital
cost multipliers for onshore wind add between 0% and 100% to the cost of a new wind plant.
3.3.1 Recommendations for Future Modeling R&D
• More robust understanding of curtailment outside current system boundaries—Current
methods for estimating VRE curtailment appear to be robust when compared with more
detailed models, but that comparison has only been made for a small number of scenarios.
Scenarios with significant changes in minimum generation levels from thermal plants,
storage deployment, or demand-side technologies could challenge the curtailment methods
that have already been developed.
• Improved methods for how financing assumptions are assigned to a technology—Financing
can have significant impacts on model results, but the financing assumptions that should
be used for a given technology are not well understood or represented in long-term planning
models. Financing parameters such as interest rate and rate of return on equity might change
over time and be especially difficult to set appropriately, but economic lifetimes can likely
be well understood and represented in the modeling.
• Improved methods for capturing VRE transmission connection costs— Transmission
connection costs can represent a significant fraction of the cost of a new plant, but it
is unclear whether current methods are underestimating or overestimating those costs and
how this might impact model results.
29
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
3.4 Other Considerations
Although the workshops focused on the topics discussed above, a number of important ancillary
topics arose. These ancillary topics were not discussed in detail, but we include them here for the
sake of completeness in documenting workshop findings and because these topics deserve ample
consideration for long-term planning R&D.
• Methods for representing energy storage—There is considerable desire to consider low-cost
storage or other similar scenarios in long-term planning models, but the methodologies and
metrics for representing storage are not yet well defined. This is especially true for the
interaction of VRE generators with storage technologies. Issues such as chronology, capacity
value, and cost representation have yet to be addressed in most large-scale modeling
frameworks.
• Improved methods for handling the demand-side—Energy efficiency, customer behavior,
electric vehicles, and general energy electrification can significantly impact electricity
demand, both in terms of total electricity demand as well as the shape of the load profile.
Methods for understanding and projecting demand-side changes at a national scale within
electricity-sector-only models are generally limited or are in their early stages of
development.
• Assessment of importance of intraregional transmission for long-term planning decisions—
Although many models capture the cost of building intraregional transmission to connect
new generators to existing transmission, none of the four models represents transmission
flows or limitations within a balancing area. It is not clear whether this is a minor assumption
for the purpose of capacity expansion or it might significantly alter model results. Also, how
this assumption might interact with selected spatial resolution is unknown.
• Metrics for assessing model behavior— Model behavior is often assessed based on expert
opinion and/or other more detailed modeling tools. Metrics that can be used to more
concretely measure and assess model behavior can accelerate model development and
improve model utility.
30
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
4 Summary and Conclusions
Capacity expansion models provide a useful tool for understanding how VRE interacts with and will
influence the future power sector under a wide range of economic, technology, and policy conditions.
However, because of the complexities associated with VRE and with long-term projections,
considerable work remains to be done to improve the utility of these models. The four modeling teams
involved in this comparison effort identified three key areas of active model enhancement to improve
the VRE representation within these planning models: (1) spatial and temporal resolution, (2) resource
adequacy, particularly the accounting method for VRE capacity, and (3) economics of energy
production, with a focus on the impact of curtailment on VRE economics.
Both the temporal and spatial resolution choice and method employed in a model can significantly
impact the results. In general we find that increased resolution leads to better representation of total
system cost and of VRE characteristics, but it is unclear how much resolution is necessary to answer
the questions asked of the model. Additionally, different questions might require different levels of
resolution.
Curtailment and capacity value of VRE are two of the most significant system characteristics to
represent in capacity expansion models. The existing methods for handling curtailment and capacity
value of VRE are generally sufficient at capturing VRE economics under systems with low-to-
moderate penetrations of VRE. However, these methods may not be sufficiently robust for futures that
are vastly different from today, and improvements will likely be needed. At a minimum, increased
understanding of high penetration VRE scenarios is needed to give increased confidence in model
solutions that show significant evolution from today’s system.
Other aspects of VRE (i.e., outside of curtailment and capacity value) do not generally have as well-
established methods, but it is also unclear to what degree they ultimately matter for ensuring projected
investment decisions are robust. For example, operating reserves, grid connection costs, and financing
costs are typically represented, but the best methods for doing so are not well defined, and the impact
of each is typically not well understood. Additionally, aspects that are typically not represented, such
as unit commitment, are also poorly understood with respect to how they might impact investment
decisions.
Across the four modeling teams that participated in our workshops, we found that most of the research
and innovation with respect to capacity expansion models utilize “out-of-optimization” methodologies.
In other words, new capabilities are developed using pre- or post-process innovations that allow the
already large and complicated optimization problem not to grow in size as additional capabilities are
added. We also found that the rapidly changing grid environment opens new challenges, especially
with respect to modeling storage and demand-side technologies.
Through this multi-model exercise, we have identified several areas for future modeling research and
development. These areas are included at the end of Sections 3.1.3, 3.2.4, 3.3.1, and 3.4. Finally, we
found that a collective environment to share ideas across modeling teams has led to higher quality
models and has bolstered individual model development, including many of the model enhancements
and method comparisons highlighted in this report (see the appendix for a summary of model
enhancements). We recommend collaboration across capacity expansion modeling teams, especially as
the research and development areas described in this report are pursued.
31
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
References
ABB. 2016. “E7 Capacity Expansion: Long-Term Resource Planning for Resource Planners and
Portfolio Managers.” http://new.abb.com/docs/librariesprovider139/default-document-
library/epm_capacity-expansion_jul16_web.pdf?sfvrsn=2.
Bird, L., J. Cochran, and X. Wang. 2014. “Wind and Solar Energy Curtailment: Experience and
Practices in the United States.” NREL/TP-6A20-60983. National Renewable Energy
Laboratory (NREL), Golden, CO. https://doi.org/10.2172/1126842.
Bistline, John. 2017. “Economic and Technical Challenges of Flexible Operations under Large-
Scale Variable Renewable Deployment.” Energy Economics 64 (Supplement C):363–72.
https://doi.org/10.1016/j.eneco.2017.04.012.
Bistline, John, and Francisco de la Chesnaye. 2017. “Banking on Banking: Does ‘When’
Flexibility Mask the Costs of Stringent Climate Policy?” Climatic Change, August, 1–14.
https://doi.org/10.1007/s10584-017-2053-6.
Bistline, John, D. Shawhan, Geoffrey Blanford, Francisco de la Chesnaye, B. Mao, Nidhi Santen,
and R. Zimmerman. 2017. “Systems Analysis in Electric Power Sector Modeling:
Evaluating Model Complexity for Long-Range Planning.” 3002011365. Palo Alto, CA:
Electric Power Research Institute.
Blanford, Geoffrey, James Merrick John Bistline, and David Young. 2016. “Simulating Annual
Variation in Load, Wind, and Solar by Representative Hour Selection.” 3002008653.
Palo Alto, CA: Electric Power Research Institute.
https://www.epri.com/#/pages/product/000000003002008653/.
Cole, Wesley, Trieu Mai, Jeffrey Logan, Daniel Steinberg, James McCall, James Richards,
Benjamin Sigrin, and Gian Porro. 2016. “2016 Standard Scenarios Report: A U.S.
Electricity Sector Outlook.” NREL/TP-6A20-66939. Golden, CO: National Renewable
Energy Laboratory. http://www.nrel.gov/docs/fy17osti/66939.pdf.
Denholm, Paul, Kara Clark, and Matt O’Connell. 2016. “On the Path to SunShot: Emerging
Issues and Challenges in Integrating High Levels of Solar into the Electrical Generation
and Transmission System.” Golden, CO: National Renewable Energy Laboratory.
http://www.osti.gov/scitech/biblio/1253978.
Denholm, Paul, and Marissa Hummon. 2012. “Simulating the Value of Concentrating Solar
Power with Thermal Energy Storage in a Production Cost Model.” NREL/TP-6A20-
56731. Golden, CO: National Renewable Energy Laboratory.
https://www.nrel.gov/docs/fy13osti/56731.pdf.
Denholm, Paul, and Robert Margolis. 2016. “Energy Storage Requirements for Achieving 50%
Solar Photovoltaic Energy Penetration in California.” NREL/TP-6A20-66595. Golden,
CO: National Renewable Energy Laboratory.
http://www.nrel.gov/docs/fy16osti/66595.pdf.
DOE. 2012. “SunShot Vision Study.” DOE/GO-102012-3037. Washington, D.C.: U.S.
Department of Energy. http://www.nrel.gov/docs/fy12osti/47927.pdfDOE/GO-102012-
3037.
Edelenbosch, O. Y., K. Kermeli, W. Crijns-Graus, E. Worrell, R. Bibas, B. Fais, S. Fujimori, P.
Kyle, F. Sano, and D. P. van Vuuren. n.d. “Comparing Projections of Industrial Energy
Demand and Greenhouse Gas Emissions in Long-Term Energy Models.” Energy.
Accessed January 26, 2017. https://doi.org/10.1016/j.energy.2017.01.017.
EIA. 2017a. “Annual Energy Outlook 2017.” DOE/EIA-0383(2017). Washington, D.C.: U.S.
DOE Energy Information Administration.
32
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
———. 2017b. “The Electricity Market Module of the National Energy Modeling System:
Model Documentation 2016.” Washington, D.C.: U.S. Energy Information
Administration.
https://www.eia.gov/outlooks/aeo/nems/documentation/electricity/pdf/m068(2016).pdf.
EPA. 2013. “Documentation for EPA Base Case v.5.13 Using the Integrated Planning Model.”
450R13002. Washington, D.C.: U.S. Environmental Protection Agency.
https://www.epa.gov/sites/production/files/2015-
07/documents/documentation_for_epa_base_case_v.5.13_using_the_integrated_planning
_model.pdf.
———. 2015a. “Carbon Pollution Emission Guidelines for Existing Stationary Sources: Electric
Utility Generating Units.” 2015. http://www.epa.gov/airquality/cpp/cpp-final-rule.pdf.
———. 2015b. “EPA Base Case v.5.15 Using IPM.” U.S. Environmental Protection Agency.
https://www.epa.gov/sites/production/files/2015-
08/documents/epa_base_case_v.5.15_incremental_documentation_august_2015.pdf.
EPRI. 2017. “US-REGEN Model Documentation.” EPRI Technical Update #3002010956. Palo
Alto, CA: Electric Power Research Institute.
Eurek, Kelly, Wesley Cole, David A. Bielen, Nate Blair, Stuart Cohen, Bethany Frew, Jonathan
Ho, Venkat Krishnan, Trieu Mai, and Daniel Steinberg. 2016. “Regional Energy
Deployment System (ReEDS) Model Documentation: Version 2016.” NREL/TP-6A20-
67067. Golden, CO: National Renewable Energy Laboratory.
http://www.nrel.gov/docs/fy17osti/67067.pdf.
Fink, Sari, C. Mudd, K. Porter, and Brett Morgenstern. 2009. “Wind Energy Curtailment Case
Studies.” NREL Subcontract Report, NREL/SR-550 46716.
http://www.nrel.gov/docs/fy10osti/46716.pdf.
Frew, Bethany A., Wesley J. Cole, Yinong Sun, Trieu T. Mai, and James Richards. 2017. “8760-
Based Method for Representing Variable Generation Capacity Value in Capacity
Expansion Models: Preprint.” NREL/CP-6A20-68869. National Renewable Energy
Laboratory (NREL), Golden, CO (United States).
https://www.osti.gov/scitech/biblio/1375305-based-method-representing-variable-
generation-capacity-value-capacity-expansion-models-preprint.
Gami, D., R. Sioshansi, and P. Denholm. 2017. “Data Challenges in Estimating the Capacity
Value of Solar Photovoltaics.” IEEE Journal of Photovoltaics 7 (4):1065–73.
https://doi.org/10.1109/JPHOTOV.2017.2695328.
Getman, Dan, Anthony Lopez, Trieu Mai, and Mark Dyson. 2015. “Methodology for Clustering
High-Resolution Spatiotemporal Solar Resource Data.” NREL/TP-6A20-63148. Golden,
CO: National Renewable Energy Laboratory.
https://www.nrel.gov/docs/fy15osti/63148.pdf.
Hale, Elaine, Brady Stoll, and Trieu Mai. 2016. “Capturing the Impact of Storage and Other
Flexible Technologies on Electric System Planning.” NREL/TP-6A20-65726. Golden,
CO: National Renewable Energy Laboratory.
http://www.nrel.gov/docs/fy16osti/65726.pdf.
Holttinen, Hannele, Juha Kiviluoma, Alain Forcione, Michael Milligan, Charles J. Smith, Jody
Dillon, Jan Dobschinski, et al. 2016. “Design and Operation of Power Systems with
Large Amounts of Wind Power: Final Summary Report, Iea Wind Task 25, Phase Three
2012-2014.” Espoo, Finland: VTT Technical Research Centre of Finland Ltd.
https://www.ieawind.org/task_25/PDF/T268.pdf.
33
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
IEA. 2015. “World Energy Model Documentation.” Paris, France: Organisation for Economic
Co-operation and Development (OECD), International Energy Agency (IEA).
Intergovernmental Panel on Climate Change (IPCC). 2015. Climate Change 2014: Mitigation of
Climate Change: Working Group III Contribution to the IPCC Fifth Assessment Report.
Cambridge: Cambridge University Press. /core/books/climate-change-2014-mitigation-
of-climate-change/81F2F8D8D234727D153EC10D428A2E6D.
Jorgensen, Jennie, Trieu Mai, and Greg Brinkman. 2017. “Reducing Wind Curtailment Through
Transmission Expansion in a Wind Vision Future.” NREL/TP--6A20-67240. Golden,
CO: National Renewable Energy Laboratory. https://doi.org/10.2172/1339078.
Jorgenson, Jennie, Paul Denholm, and Mark Mehos. 2014. “Estimating the Value of Utility-
Scale Solar Technologies in California Under a 40% Renewable Portfolio Standard.”
NREL/TP-6A20-61685. Golden, CO: National Renewable Energy Laboratory.
https://doi.org/10.2172/1134134.
Krishnan, Venkat, and Wesley Cole. 2016. “Evaluating the Value of High Spatial Resolution in
National Capacity Expansion Models Using Reeds.” Poster NREL/PO-6A20-66786.
Golden, CO: National Renewable Energy Laboratory.
http://www.nrel.gov/docs/fy16osti/66002.pdf.
Kroposki, B., B. Johnson, Y. Zhang, V. Gevorgian, P. Denholm, B. M. Hodge, and B. Hannegan.
2017. “Achieving a 100% Renewable Grid: Operating Electric Power Systems with
Extremely High Levels of Variable Renewable Energy.” IEEE Power and Energy
Magazine 15 (2):61–73. https://doi.org/10.1109/MPE.2016.2637122.
Mai, Trieu, Clayton Barrows, Anthony Lopez, Elaine Hale, Mark Dyson, and Kelly Eurek. 2015.
“Implications of Model Structure and Detail for Utility Planning: Scenario Case Studies
Using the Resource Planning Model.” NREL/TP--6A20-63972. National Renewable
Energy Lab. (NREL), Golden, CO (United States).
https://www.osti.gov/scitech/biblio/1334388-implications-model-structure-detail-utility-
planning-scenario-case-studies-using-resource-planning-model.
Mai, Trieu, John Bistline, Yinong Sun, Cara Marcy, Christopher Namovicz, David Young, and
Wesley Cole. forthcoming. “Insights into the Future of Variable Renewable Energy: A
Multi-Model Perspective of the U.S. Electricity System.”
Mai, Trieu, R. Wiser, D. Sandor, G. Brinkman, G. Heath, P. Denholm, D. J. Hostick, N.
Darghouth, A. Schlosser, and K. Strzepek. 2012. “Exploration of High-Penetration
Renewable Electricity Futures. Vol. 1 of Renewable Electricity Futures Study.”
NREL/TP-6A20-52409-1. Golden, CO: National Renewable Energy Laboratory.
Milligan, Michael, Bethany A. Frew, Aaron Bloom, Erik Ela, Audun Botterud, Aaron Townsend,
and Todd Levin. 2016. “Wholesale Electricity Market Design with Increasing Levels of
Renewable Generation: Revenue Sufficiency and Long-Term Reliability.” The Electricity
Journal 29 (2):26–38. https://doi.org/10.1016/j.tej.2016.02.005.
Mills, Andrew, and Ryan Wiser. 2012. “Changes in the Economic Value of Variable Generation
at High Penetration Levels: A Pilot Case Study of California.” LBNL-5445E. Berkeley,
CA: Lawrence Berkely National Laboratory. http://emp.lbl.gov/sites/all/files/lbnl-
5445e.pdf.
Nahmmacher, Paul, Eva Schmid, Lion Hirth, and Brigitte Knopf. 2016. “Carpe Diem: A Novel
Approach to Select Representative Days for Long-Term Power System Modeling.”
Energy 112 (Supplement C):430–42. https://doi.org/10.1016/j.energy.2016.06.081.
34
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Nelson, James, Josiah Johnston, Ana Mileva, Matthias Fripp, Ian Hoffman, Autumn Petros-
Good, Christian Blanco, and Daniel M. Kammen. 2012. “High-Resolution Modeling of
the Western North American Power System Demonstrates Low-Cost and Low-Carbon
Futures.” Energy Policy 43 (Supplement C):436–47.
https://doi.org/10.1016/j.enpol.2012.01.031.
NERC. 2017. “2017 Summer Reliability Assessment.”
http://www.nerc.com/pa/RAPA/ra/Reliability%20Assessments%20DL/2017%20Summer
%20Assessment.pdf.
NREL. 2016. “2016 Annual Technology Baseline.” Golden, CO: National Renewable Energy
Laboratory. http://www.nrel.gov/analysis/data_tech_baseline.html.
———. 2017. “2017 Annual Technology Baseline.” Golden, CO: National Renewable Energy
Laboratory. http://www.nrel.gov/analysis/data_tech_baseline.html.
Pietzcker, Robert C., Falko Ueckerdt, Samuel Carrara, Harmen Sytze de Boer, Jacques Després,
Shinichiro Fujimori, Nils Johnson, et al. n.d. “System Integration of Wind and Solar
Power in Integrated Assessment Models: A Cross-Model Evaluation of New
Approaches.” Energy Economics. Accessed January 25, 2017.
https://doi.org/10.1016/j.eneco.2016.11.018.
Pietzcker, Robert Carl, Daniel Stetter, Susanne Manger, and Gunnar Luderer. 2014. “Using the
Sun to Decarbonize the Power Sector: The Economic Potential of Photovoltaics and
Concentrating Solar Power.” Applied Energy 135 (December):704–20.
https://doi.org/10.1016/j.apenergy.2014.08.011.
Santen, Nidhi, John Bistline, Geoffrey Blanford, and Francisco de la Chesnaye. 2017. “Electric
Power Sector Systems Analysis: A Review of the Recent Literature and Capabilities of
Selected Capacity Planning Tools.” 3002011102. Palo Alto, CA: Electric Power
Research Institute. https://www.epri.com/#/pages/summary/000000003002011102/.
Sigrin, B., P. Sullivan, E. Ibanez, and R. Margolis. 2014. “Representation of the Solar Capacity
Value in the Reeds Capacity Expansion Model: Preprint.” NREL/CP-6A20-62015.
National Renewable Energy Laboratory (NREL), Golden, CO.
https://www.osti.gov/scitech/biblio/1150799-representation-solar-capacity-value-reeds-
capacity-expansion-model-preprint.
Steinberg, Daniel, Dave Bielen, Josh Eichman, Kelly Eurek, Jeff Logan, Trieu Mai, Colin
McMillan, Andrew Parker, Laura Vimmerstedt, and Eric Wilson. 2017. “Electrification
and Decarbonization: Exploring U.S. Energy Use and Greenhouse Gas Emissions in
Scenarios with Widespread Electrification and Power Sector Decarbonization.”
NREL/TP--6A20-68214. Golden, CO: National Renewable Energy Laboratory.
https://doi.org/10.2172/1372620.
Sun, Yinong, and Wesley J. Cole. 2017. “Transmission Flow Methodologies: Approximate DC
Flow vs. Pipe Flow Along AC Lines.” NREL/PR-6A20-68929. Golden, CO: National
Renewable Energy Laboratory. https://www.nrel.gov/docs/fy17osti/68929.pdf.
Ueckerdt, Falko, Lion Hirth, Gunnar Luderer, and Ottmar Edenhofer. 2013. “System LCOE:
What Are the Costs of Variable Renewables?” Energy 63 (Supplement C):61–75.
https://doi.org/10.1016/j.energy.2013.10.072.
WECC. 2013. “2013 Interconnection-Wide Plan Tools and Models.”
https://www.wecc.biz/Reliability/2013Plan_ToolsandModels.docx
35
This report is available at no cost from the National Renewable Energy Laboratory at www.nrel.gov/publications.
Appendix. Summary of Model Enhancements
One purpose of the two CEM modeling workshops was to stimulate improvements and examine
assumptions in each the models, especially as they relate to VRE representation. Each of the
modeling teams identified and/or incorporated improvements and analyses between the two
workshops, and a summary of those is provided here.
IPM
The EPA IPM team is planning to implement several improvements to VRE representation in
the upcoming release of IPM v6. The most significant improvements include implementation
of 8760 hourly generation profiles, increased time slices targeting more resolution during
baseload hours, time-of-day differentiation, an additional winter season to better capture
variability in wind performance, VRE spur line costs integrated with the capital costs of new
units, and a new methodology to impose declining capacity value on new VRE resources.
NEMS
EIA has scheduled several modeling updates focused on improving representation of the value
and impact of VRE generation on the grid. Between AEO 2016 and the results shared in this
report, EIA updates to NEMS included changing the representation of residential solar PV from
a simple cash flow model to a statistical model, capturing a more detailed representation of the
end-use solar PV generation on net demand, and modeling solar PV and wind curtailments. In
addition, for AEO 2018, other modeling updates include integrating a four-hour energy storage
technology and modifying the regional solar resource to reflect a supply curve rather than an
average value. In addition to model updates, EIA conducted a number of model experiments,
including an analysis of the curtailment algorithm and a review of the spinning reserve
algorithm.
ReEDS
The most significant change to the ReEDS model was the incorporation of a new VRE capacity
value calculation methodology that uses hourly data to assess the contribution of existing and
potential new VRE generators toward the resource adequacy requirement. The methodology was
discussed briefly in Section 3.1.1, and details are reported by Frew et al. (2017). In addition, the
ReEDS team assessed the difference between a pipe-flow transmission representation and a DC-
power-flow representation (Sun and Cole 2017), and it did model runs comparing different
financing assumptions Finally, the incremental value used for calculating marginal VRE capacity
values and curtailment rates was examined to see how it impacts model results.
US-REGEN
The US-REGEN team conducted experiments to explore the impacts of spatial and temporal
resolution on model outputs. Detailed information about these analyses, results, and implications
for modeling can be found in Bistline et al. (2017). The team also added a wider variety of
utility-scale PV technologies, and the model now allows investments in fixed-tilt, single-axis
tracking, and double-axis tracking technologies. In addition, EPRI updated its representations
of operating reserves, planning reserves, and energy storage to illustrate potential effects of
alternate formulations and parameters.
