3

Models for Long-Term Planning

The workshop’s first session focused on models for long-term planning. Karen Palmer, Resources for the Future, introduced the speakers: David Daniels, U.S. Energy Information Administration (EIA); Daniel Shawhan, Resources for the Future; John Bistline, Electric Power Research Institute (EPRI); Bethany Frew, National Renewable Energy Laboratory (NREL); and John Larsen, Rhodium Group.

Palmer cautioned that energy models do not predict the future, but they are nonetheless useful. “They’re not a crystal ball,” she said. “Models are a useful way to represent complex systems, tweak inputs, and gain insights.” While energy predictions for future decades can be imprecise, by continually improving models’ inputs and assumptions, it is possible to further improve their usefulness. She added that energy modeling has never been easy, but it is particularly challenging today because of the energy sector’s unprecedented pace of change.

In their remarks and the discussion that followed, speakers discussed how models are used to understand long-term changes and needs in the electric power system.

DAVID DANIELS, U.S. ENERGY INFORMATION ADMINISTRATION

Daniels, chief energy modeler at the EIA, addressed the use of modeling as a tool to help policy makers understand how their decisions impact the energy sector. He cautioned that models should not be seen as tools for predicting the future—instead, they are more like calculators that

simulate the effects of decisions. “They’re just fancy calculators,” he said. “They don’t replicate the decisions of the real world, … [where] people are making decisions for all kinds of reasons, competitive pressures, all kinds of things. But models do simulate the effects of these decisions in the real world.”

The EIA’s Annual Energy Outlook, for example, includes a reference case and various results based on different scenarios. If the price of renewable energy goes down, the model shows a rise in renewables use, but a slower rise in renewables if the supply of oil and gas should increase. By producing these simulations under an assumption that current laws and regulations remain unchanged, the model provides Congress with a baseline to show how policy changes could potentially lead to desired outcomes.

Daniels stated that the real world is far too large and complex for a model to fully represent. Instead, modelers make simplified representations of certain aspects that they want to focus on. In doing so, they must make trade-offs between breadth and depth across multiple dimensions—such as region, time, sector, and fuel/technology—and the resulting models represent only a part of the multidimensional phase space of the real world, much like a telescope can be pointed at only one part of the sky.

Because all models have such limits, no single model can fully answer a given question. Together, however, intermodel comparisons could come close, and Daniels suggested that policy ideas be modeled on multiple platforms looking at different areas of phase space. “Individually, no model can tell you the answer,” he said. “Collectively, though, if you get an ensemble of models that are heterogeneous enough, you can cover the whole phase space.”

Another concept Daniels mentioned was policy robustness, checking that the outcomes of a policy were consistent across different input assumptions. Ideally, one should explore the future impacts of a policy with an ensemble of heterogeneous models across a range of potential future scenarios, he said.

For modeling future power systems in particular, he said it is important that models distinguish between cost and value, allowing them to anticipate future markets that effectively compensate for value provided.

DANIEL SHAWHAN, RESOURCES FOR THE FUTURE

Shawhan, fellow at Resources for the Future and adjunct assistant professor at Cornell, spoke about his experience developing and using the Engineering, Economic, and Environmental Electricity Simulation Tool (or “E4ST,” described at E4ST.org). E4ST projects market outcomes by combining the decisions of energy consumers and investor-owners

for maximum consumer benefit and minimum owner cost, within the existing transmission constraints. To test model accuracy, E4ST has back-simulated 2013, reproducing observed power market outcomes well.

E4ST was designed with a high degree of detail and realism, incorporating the thousands of U.S. and Canadian high-voltage transmission lines, a simplified representation of the low-voltage transmission lines, and all of the grid-serving generators in order to project the impacts of policy changes, infrastructure investments, and other factors (Figure 3.1). It includes endogenous generator investment and retirement; physics-based transmission flows; price-responsive demand at each node; comprehensive cost-benefit analyses; and, unusually for an energy model, air pollution and its health effects.

Shawhan made several suggestions. First, he suggested that power-sector policy analysis and planning should use comprehensive benefit-cost analysis, including the estimated value of the effects on users, sellers, taxpayers, health, and the environment. This can be further improved upon by estimating general equilibrium effects, distributional effects, and reliability and resilience effects, he noted.

He underscored the importance of air pollution modeling for power sector decisions. To improve the estimation of environmental effects of investment, operational, and policy decisions, the most available air pollution models should be improved, and the best air pollution models should be made more available and less costly for power sector modelers to use, he said.

If governments charged power plants for the estimated damage from their emissions, including both greenhouse gases and local and regional pollutants, that would save thousands of lives per year in the United States and lower the overall total cost of electricity borne by society, according to Shawhan’s model results (Figure 3.2). Until then, Shawhan suggested that decisions subject to regulation should be required to consider estimated air pollution benefits and costs on an equal basis with other benefits and costs.

Shawhan noted that all power sector policy analysis and planning models, even E4ST, have room for improvement in terms of their realism. There is much that can be done to improve the extent to which models can incorporate additional important phenomena. He argued that it should be possible to construct a model that can estimate and value the environmental, reliability, resilience, and pocketbook effects of system investments all at the same time.

Last, Shawhan suggested that making distribution charges match short-run marginal cost more closely, as has been done with transmission charges, would greatly reduce distribution costs, promote renewables, and reduce inefficient grid defection. If done optimally, this dynamic distribution pricing would create a shortfall of distribution revenue, but that shortfall could be covered by revenue from charges on environmental damages. The result of this combination would be a large emission reduction with little or no increase in electricity bills, he said.

JOHN BISTLINE, ELECTRIC POWER RESEARCH INSTITUTE

Bistline, principal technical leader in the Energy Systems and Climate Analysis Group at EPRI, detailed the U.S. Regional Economy, Greenhouse Gas, and Energy (US-REGEN) model, EPRI’s in-house model for economic insights, power sector decision support, and understanding the impacts of policy and technological changes at the state, national, and international levels.

US-REGEN includes enormous detail and regional heterogeneity, which are necessary to understand economic and behavioral incentives for companies and households. The end-use portion of the model accounts for a range of drivers and impacts across buildings, transportation, and industrial sectors. “Our goal when we built this part of US-REGEN a few years ago was to have as much detail on the demand side as we had on the supply side,” said Bistline. “And so I think our modeling is unique in that we’re capturing the economic and behavioral incentives of different types of households, instead of just assuming a particular deployment of technologies over time.”

In addition, US-REGEN synchronizes solutions between end-use and electric system models, iterating scenarios with detailed, hourly sectoral load changes until the prices converge. The model is quite robust, having passed through careful scrutiny and detailed vetting from a variety of experts and industry stakeholders, Bistline noted.

US-REGEN’s most prominent features are its treatment of variable renewables, energy storage, and end-use loads. US-REGEN can also dis-aggregate information to understand hourly and seasonal load shapes, asset utilization, flexibility needs, and research and development (R&D) opportunities (Figure 3.3). It also incorporates variation in renewables generation—for example, whether there is high or low wind or solar output—and can model hourly energy use, which is especially important for energy storage and can vary depending on the region and season. “Thinking about spatial and temporal resolution is really important for a lot of resources, but especially [for] variable renewables like wind and solar,” Bistline said. “That’s something that we worked hard to characterize.”

Bistline concluded by sharing US-REGEN’s planned future improvements, including increased spatial and temporal granularity, linkages with reliability and production cost models, the incorporation of low-carbon fuel pathways such as hydrogen, and use of decomposition approaches to solve more detailed models.

BETHANY FREW, NATIONAL RENEWABLE ENERGY LABORATORY

Frew, senior researcher in the Economics and Forecasting Group of the Strategic Energy Analysis Center at NREL, began by showing an overview of NREL’s existing power system modeling capabilities, which cover a wide range of spatial and temporal resolutions (Figure 3.4). Much effort has been taken to manually build linkages between individual models to provide insights on interactions across areas that have traditionally been more siloed, Frew said.

She then discussed two specific capacity-expansion models being developed or already used at NREL. These are the flagship Regional Energy Deployment System (ReEDS) model and a new electricity market design test-bed called Electricity Markets and Investment Suite (EMIS). Within ReEDS, significant effort is being devoted to better represent the challenges of variable renewable energy integration—for example, by increasing temporal and spatial resolution and developing a more detailed representation of storage. ReEDS, which is now open access, was recently updated to include improvements such as flexible solve-structure, demand-side representation, and coverage of all of North America. EMIS is currently being developed to evaluate the impact of market design and heterogeneous investor firms on investment decisions and reliability.

EMIS will be able to evaluate how various market designs, which can be customized by the end user, perform under uncertainty.

Frew said that NREL is also developing a broader modeling framework, Scalable Integrated Infrastructure Planning (SIIP), to which EMIS will belong. To better streamline workflow and improve cross-model linkages, SIIP’s unifying modeling framework is being used to overcome challenges related to differences between software languages, inconsistent data structures, and the inability to co-optimize multiple models (Figure 3.5).

Furthermore, EMIS is being developed as a suite of three different model formulations that can inform decision making from different angles. “It’s not just one formulation,” Frew said, explaining that developers aim to co-optimize firm-level investment decisions with the market clearing process across multiple years and possible futures. This will enable investment decisions to be made based on an expectation of the future market revenues for various combinations of market products and future outcomes, she said. For example, EMIS could be used to evaluate the impact of market design evolution on grid buildout and reliability.

JOHN LARSEN, RHODIUM GROUP

John Larsen is the director of U.S. Energy Research at Rhodium Group, which uses modeling to answer questions relevant to the future of the electric power system. His particular focus is economy-wide deep decarbonization pathways such as energy efficiency, electrification, and carbon dioxide removal (e.g., through direct air capture [DAC]), all of which shape load demand and energy economics. He discussed how the modeling community can inform efforts to move toward net zero emissions.

Rhodium Group uses two main models to study deep decarbonization. The first is a modification of the EIA’s National Energy Modeling System (NEMS) called RHG-NEMS, which includes greenhouse gas projections and provides a more comprehensive assessment of factors such as cost and performance than NEMS does alone. This tool is primarily used for projecting near-term trends in the range of 10-15 years. The second is a combination of two models: Regional Investment and Operations (RIO)

Platform and Energy Pathways, owned and operated by Evolved Energy Research, which can account for a wider range of future technology, cost, and performance changes and produce demand projections and hour-by-hour representations of the electric system. Each platform has strengths and weaknesses for applications in decarbonization analysis.

Larsen shared some projections from Rhodium Group’s use of these models. RHG-NEMS projections suggest that increased carbon prices can reduce carbon emissions (Figure 3.6), but not achieve deep decarbonization—for that, additional technology deployment in end-use sectors and carbon removal such as DAC would be required (Figure 3.7). Carbon removal is not explicitly characterized in RHG-NEMS, but Rhodium’s cost estimates of the technology suggest that DAC could become economically attractive in the next 10-20 years as the cost of these technologies drops and carbon prices rise. Researchers also used RHG-NEMS and other resources to compare cost performance for electrofuels and fossil fuels as their prices rise and fall. Rhodium’s analysis shows that direct carbon capture technologies may become economically attractive.

In addition, Rhodium Group analyses suggest that electrofuel use could grow as carbon prices rise, despite being cost-prohibitive today. Electrification of vehicles and other products is also projected to rise, although not without better battery costs and aggressive policy interventions, which Larsen suggested will make electrofuels a nontrivial amount

of the energy supply by 2050. While direct carbon capture and electrofuels require high start-up costs and high utilization rates to be economical, Rhodium Group’s modeling suggests scenarios where they can play a substantial role in a decarbonized energy system. If these technologies do succeed in ramping up, they will have material impacts on the size and shape of electric demand in a deep decarbonized energy system, Larsen said.

DISCUSSION

Following the speakers’ remarks, Palmer moderated an open discussion period. Participants considered the role of resiliency, modeling for decision makers, opportunities for improvements, and the international context.

Resiliency in Models

J.P. Watson, Lawrence Livermore National Laboratory (LLNL), asked why resiliency objectives do not appear to be a prominent focus in modeling. Shawhan answered that resilience is both difficult and expensive to measure and incorporate. There are also strong differences of opinion regarding how to address resiliency, particularly between funders and

researchers. Building on this point, Bistline noted that resiliency means different things to different stakeholders. For example, hotter summers impact peak load and resource planning, but experts disagree on whether this falls under the realm of resiliency planning.

Frew added that resiliency questions are challenging based on data, software model, and metric needs, where there may not be consistent definitions or data to properly characterize low-probability, high-impact resilience events. NREL does resilience work to support cities and regions in assessing and responding to extreme events; these efforts utilize many of the modeling linkages that Frew described during her talk (e.g., capacity expansion to determine a system buildout, production cost modeling for operations, and probabilistic resource adequacy to determine times and locations of potential dropped load). Frew also noted that NREL is working to develop next-generation model capabilities and also to acquire richer data sets to better characterize extreme weather or climate scenarios—for example, in a recent data set of multiple years of load data under different climate change scenarios.

Modeling for Decision Makers

Susan Tierney, Analysis Group, asked the speakers how modeling could be more helpful for informing decision making. Daniels answered that it is difficult for policy makers to understand the nuances of models and what they can and cannot provide. For instance, while models do not predict the future, they can estimate how actions taken now could change the future. Given the nonlinearities in the real world and the boundary effects of simplified representations in models, trying to model large changes could be misleading; instead, showing decision makers the impacts of a series of successively larger, but still incremental changes (e.g., a series of small, but increasing carbon prices) could more effectively contribute to the decision-making process, he suggested.

Another participant suggested that decision makers be invited to an interactive exercise that allows them to adjust a model’s parameters and see the results change, thus learning through experience how the model shows the impact of a policy change on emissions or costs. The FAST Predictor version of E4ST is one example of this approach.

Bistline noted that EPRI has taken a proactive approach to communication, including holding public seminars for Capitol Hill staff members, researchers, and other stakeholders to summarize what they have learned from modeling. “Making sure you allocate that effort and time to do something like that at the end is really, really valuable,” he said.

Another participant added that making models more relevant to decision makers will require greater appreciation for the fact that models by

necessity work in generalities. Decision makers often seek answers to very specific questions, while models are generally better suited to elucidating how a small change affects the larger picture. “What they want to know … sounds like this: ‘What happens in my little area that I care about deeply, when I change everything?’ And unfortunately, the answer is, ‘We can’t tell you.’” Instead, the participant suggested that modeling as a whole is better positioned to support perturbation analyses, such as those that integrated assessment models (IAMs) are designed to perform, in which the focus is on large-scale global and economic impacts from tiny changes in a small number of parameters.

Model Improvements and Validations

Asked for suggestions on ways to improve modeling, Daniels answered that it is very important for platforms to be flexible, so that they can easily swap different policies, market changes, or technologies. The world is rapidly changing, and the ideal platform would accommodate that change, he said.

Another participant asked how the models discussed today are validated, and if improvements are needed. Frew answered that model comparison activities are highly valuable, and NREL has had success conducting those in the past, resulting in stronger tools. Bistline, a member of NREL’s comparison group, agreed. Another participant added that using models to forecast past events is an effective validation practice.

International Context

A participant raised the international context of large-scale decarbonization: If one country takes action to dramatically decarbonize but the rest of the world does not, how does this affect the costs and benefits borne by different countries and economies? Shawhan answered that models can represent some such international effects, such as effects on generation technology costs, fuel costs, air pollution, and climate. Global models offer some insight, but they lack the technical detail necessary to represent power sectors realistically, and it is important to foster open dialogue and share best practices to align needs with models, he said. Frew added that there are several European IAMs with this goal.

A participant pointed out that modeling will have difficulty creating accurate outcomes given a global scope and most nations’ and industries’ “zero-sum” competitive mentality. Power, transportation, building, and heavy industry are interconnected and competitive, and those relationships should be taken into account when possible, the participant said.